Deleting autotokenizer
Browse files- config_utils.py +0 -769
- general_utils.py +0 -309
- sequtils.py +0 -980
- tokenizer.py +0 -363
- tokenizer_config.json +1 -7
config_utils.py
DELETED
@@ -1,769 +0,0 @@
|
|
1 |
-
# Config utils
|
2 |
-
import yaml
|
3 |
-
import pathlib
|
4 |
-
from os.path import join
|
5 |
-
import os
|
6 |
-
import numpy as np
|
7 |
-
import torch
|
8 |
-
import argparse
|
9 |
-
from multiprocessing import cpu_count
|
10 |
-
from transformers import TrainingArguments
|
11 |
-
from copy import deepcopy
|
12 |
-
import re
|
13 |
-
import sys
|
14 |
-
|
15 |
-
def add_hf_args_to_parser(parser):
|
16 |
-
# Create a temporary TrainingArguments to access default values and descriptions
|
17 |
-
hf_args = TrainingArguments(output_dir="/tmp") # Dummy output_dir
|
18 |
-
# Iterate over all public attributes
|
19 |
-
for attr in dir(hf_args):
|
20 |
-
if not attr.startswith("_"):
|
21 |
-
default = getattr(hf_args, attr)
|
22 |
-
# You can add more sophisticated handling based on attribute types here
|
23 |
-
if isinstance(default, (int, float, str, bool)):
|
24 |
-
help_str = f"Auto-generated help for {attr}"
|
25 |
-
parser.add_argument(f"--{attr}", type=type(default), default=default, help=help_str)
|
26 |
-
|
27 |
-
return parser
|
28 |
-
|
29 |
-
class BaseConfig:
|
30 |
-
"""Base class for managing and validating configurations."""
|
31 |
-
|
32 |
-
numpy_dtype_mapping = {1: np.int8,
|
33 |
-
2: np.int16,
|
34 |
-
8: np.int64,
|
35 |
-
4: np.int32}
|
36 |
-
|
37 |
-
def __init__(self):
|
38 |
-
super().__init__()
|
39 |
-
|
40 |
-
def cast_to_expected_type(self, parameter_class: str, parameter_name: str, value: any) -> any:
|
41 |
-
"""
|
42 |
-
Cast the given value to the expected type.
|
43 |
-
|
44 |
-
:param parameter_class: The class/category of the parameter.
|
45 |
-
:type parameter_class: str
|
46 |
-
:param parameter_name: The name of the parameter.
|
47 |
-
:type parameter_name: str
|
48 |
-
:param value: The value to be casted.
|
49 |
-
:type value: any
|
50 |
-
:return: Value casted to the expected type.
|
51 |
-
:rtype: any
|
52 |
-
:raises ValueError: If casting fails.
|
53 |
-
"""
|
54 |
-
expected_type = self.parameters[parameter_class][parameter_name]['type']
|
55 |
-
|
56 |
-
if expected_type in ["integer", "int"]:
|
57 |
-
try:
|
58 |
-
return int(value)
|
59 |
-
except ValueError:
|
60 |
-
raise ValueError(f"Failed to cast value '{value}' to integer for parameter '{parameter_name}' in class '{parameter_class}'.")
|
61 |
-
elif expected_type == "float":
|
62 |
-
try:
|
63 |
-
return float(value)
|
64 |
-
except ValueError:
|
65 |
-
raise ValueError(f"Failed to cast value '{value}' to float for parameter '{parameter_name}' in class '{parameter_class}'.")
|
66 |
-
elif expected_type in ["string", "str"]:
|
67 |
-
return str(value)
|
68 |
-
elif expected_type in ["boolean", "bool"]:
|
69 |
-
if isinstance(value, bool):
|
70 |
-
return value
|
71 |
-
elif str(value).lower() == "true":
|
72 |
-
return True
|
73 |
-
elif str(value).lower() == "false":
|
74 |
-
return False
|
75 |
-
else:
|
76 |
-
raise ValueError(f"Failed to cast value '{value}' to boolean for parameter '{parameter_name}' in class '{parameter_class}'.")
|
77 |
-
elif expected_type == "type":
|
78 |
-
# For this type, we will simply return the value without casting.
|
79 |
-
# It assumes the configuration provides valid Python types.
|
80 |
-
return value
|
81 |
-
elif expected_type == "list":
|
82 |
-
if isinstance(value, list):
|
83 |
-
return value
|
84 |
-
else:
|
85 |
-
raise ValueError(f"Failed to validate value '{value}' as a list for parameter '{parameter_name}' in class '{parameter_class}'.")
|
86 |
-
elif expected_type == "tuple":
|
87 |
-
if isinstance(value, tuple):
|
88 |
-
return value
|
89 |
-
else:
|
90 |
-
raise ValueError(f"Failed to validate value '{value}' as a tuple for parameter '{parameter_name}' in class '{parameter_class}'.")
|
91 |
-
elif expected_type == "set":
|
92 |
-
if isinstance(value, set):
|
93 |
-
return value
|
94 |
-
else:
|
95 |
-
raise ValueError(f"Failed to validate value '{value}' as a set for parameter '{parameter_name}' in class '{parameter_class}'.")
|
96 |
-
elif expected_type == "dict":
|
97 |
-
if isinstance(value, dict):
|
98 |
-
return value
|
99 |
-
else:
|
100 |
-
raise ValueError(f"Failed to validate value '{value}' as a dict for parameter '{parameter_name}' in class '{parameter_class}'.")
|
101 |
-
else:
|
102 |
-
raise ValueError(f"Unknown expected type '{expected_type}' for parameter '{parameter_name}' in class '{parameter_class}'.")
|
103 |
-
|
104 |
-
|
105 |
-
|
106 |
-
def get_parameter(self, parameter_class: str, parameter_name: str) -> any:
|
107 |
-
"""
|
108 |
-
Retrieve the default value of a specified parameter.
|
109 |
-
|
110 |
-
:param parameter_class: The class/category of the parameter (e.g., 'segmentation').
|
111 |
-
:type parameter_class: str
|
112 |
-
:param parameter_name: The name of the parameter.
|
113 |
-
:type parameter_name: str
|
114 |
-
:return: Default value of the parameter, casted to the expected type.
|
115 |
-
:rtype: any
|
116 |
-
"""
|
117 |
-
default_value = self.parameters[parameter_class][parameter_name]['default']
|
118 |
-
return self.cast_to_expected_type(parameter_class, parameter_name, default_value)
|
119 |
-
|
120 |
-
|
121 |
-
|
122 |
-
def validate_type(self, parameter_class: str, parameter_name: str, value: any) -> bool:
|
123 |
-
"""
|
124 |
-
Validate the type of a given value against the expected type.
|
125 |
-
|
126 |
-
:param parameter_class: The class/category of the parameter.
|
127 |
-
:type parameter_class: str
|
128 |
-
:param parameter_name: The name of the parameter.
|
129 |
-
:type parameter_name: str
|
130 |
-
:param value: The value to be validated.
|
131 |
-
:type value: any
|
132 |
-
:return: True if the value is of the expected type, otherwise False.
|
133 |
-
:rtype: bool
|
134 |
-
"""
|
135 |
-
expected_type = self.parameters[parameter_class][parameter_name]['type']
|
136 |
-
|
137 |
-
if expected_type == "integer" and not isinstance(value, int):
|
138 |
-
return False
|
139 |
-
elif expected_type == "float" and not isinstance(value, float):
|
140 |
-
return False
|
141 |
-
elif expected_type == "string" and not isinstance(value, str):
|
142 |
-
return False
|
143 |
-
else:
|
144 |
-
return True
|
145 |
-
|
146 |
-
def validate_value(self, parameter_class: str, parameter_name: str, value: any) -> bool:
|
147 |
-
"""
|
148 |
-
Validate the value of a parameter against its constraints.
|
149 |
-
|
150 |
-
:param parameter_class: The class/category of the parameter.
|
151 |
-
:type parameter_class: str
|
152 |
-
:param parameter_name: The name of the parameter.
|
153 |
-
:type parameter_name: str
|
154 |
-
:param value: The value to be validated.
|
155 |
-
:type value: any
|
156 |
-
:return: True if the value meets the constraints, otherwise False.
|
157 |
-
:rtype: bool
|
158 |
-
"""
|
159 |
-
constraints = self.parameters[parameter_class][parameter_name].get('constraints', {})
|
160 |
-
|
161 |
-
if 'options' in constraints and value not in constraints['options']:
|
162 |
-
return False
|
163 |
-
if 'min' in constraints and value < constraints['min']:
|
164 |
-
return False
|
165 |
-
if 'max' in constraints and value > constraints['max']:
|
166 |
-
return False
|
167 |
-
return True
|
168 |
-
|
169 |
-
|
170 |
-
def validate(self, parameter_class: str, parameter_name: str, value: any):
|
171 |
-
"""
|
172 |
-
Validate both the type and value of a parameter.
|
173 |
-
|
174 |
-
:param parameter_class: The class/category of the parameter.
|
175 |
-
:type parameter_class: str
|
176 |
-
:param parameter_name: The name of the parameter.
|
177 |
-
:type parameter_name: str
|
178 |
-
:param value: The value to be validated.
|
179 |
-
:type value: any
|
180 |
-
:raises TypeError: If the value is not of the expected type.
|
181 |
-
:raises ValueError: If the value does not meet the parameter's constraints.
|
182 |
-
"""
|
183 |
-
if not self.validate_type(parameter_class, parameter_name, value):
|
184 |
-
raise TypeError(f"Invalid type for {parameter_name} for parameter class '{parameter_class}'. Expected {self.parameters[parameter_class][parameter_name]['type']}.")
|
185 |
-
|
186 |
-
if not self.validate_value(parameter_class, parameter_name, value):
|
187 |
-
raise ValueError(f"Invalid value for {parameter_name} for parameter class '{parameter_class}'. Constraints: {self.parameters[parameter_class][parameter_name].get('constraints', {})}.")
|
188 |
-
|
189 |
-
def describe(self, parameter_class: str, parameter_name: str) -> str:
|
190 |
-
"""
|
191 |
-
Retrieve the description of a parameter.
|
192 |
-
|
193 |
-
:param parameter_class: The class/category of the parameter.
|
194 |
-
:type parameter_class: str
|
195 |
-
:param parameter_name: The name of the parameter.
|
196 |
-
:type parameter_name: str
|
197 |
-
:return: Description of the parameter.
|
198 |
-
:rtype: str
|
199 |
-
"""
|
200 |
-
return self.parameters[parameter_class][parameter_name]['description']
|
201 |
-
|
202 |
-
@staticmethod
|
203 |
-
def rename_non_unique_parameters(config: dict) -> tuple[dict, dict, dict]:
|
204 |
-
"""
|
205 |
-
Rename parameters in the configuration to ensure uniqueness across different groups.
|
206 |
-
|
207 |
-
This method identifies parameters with the same name across different groups and renames them
|
208 |
-
by prefixing the group name. This is to prevent conflicts when parameters are used in a context
|
209 |
-
where the group name is not specified.
|
210 |
-
|
211 |
-
:param config: A dictionary where each key is a group name and each value is a dict
|
212 |
-
of parameters for that group.
|
213 |
-
:type config: dict
|
214 |
-
|
215 |
-
:return: A tuple containing:
|
216 |
-
- renamed_config: A dictionary with the same structure as the input, but with non-unique parameter
|
217 |
-
names renamed. The structure is {group_name: {param_name: param_info}}.
|
218 |
-
- cmd_argument2group_param: A dictionary mapping the new parameter names to their original group
|
219 |
-
and parameter name. The structure is {new_param_name: [group_name, original_param_name]}.
|
220 |
-
- group2param2cmdarg: A dictionary mapping each group to a dict that maps the original parameter
|
221 |
-
names to the new parameter names. The structure is {group_name: {original_param_name: new_param_name}}.
|
222 |
-
:rtype: tuple[dict, dict, dict]
|
223 |
-
"""
|
224 |
-
|
225 |
-
# Identify non-unique parameter names
|
226 |
-
param_counts = {}
|
227 |
-
for group_name, parameters in config.items():
|
228 |
-
for param_name in parameters.keys():
|
229 |
-
param_counts[param_name] = param_counts.get(param_name, 0) + 1
|
230 |
-
|
231 |
-
non_unique_params = {param for param, count in param_counts.items() if count > 1}
|
232 |
-
|
233 |
-
cmd_argument2group_param = {}
|
234 |
-
group2param2cmdarg = {}
|
235 |
-
for group_name, parameters in config.items():
|
236 |
-
group2param2cmdarg[group_name]={}
|
237 |
-
for param_name in parameters.keys():
|
238 |
-
group2param2cmdarg[group_name][param_name] = param_name
|
239 |
-
|
240 |
-
|
241 |
-
# Rename only the non-unique parameters
|
242 |
-
renamed_config = {}
|
243 |
-
for group_name, parameters in config.items():
|
244 |
-
renamed_group = {}
|
245 |
-
for param_name, param_info in parameters.items():
|
246 |
-
|
247 |
-
new_param_name = f"{group_name}_{param_name}" if param_name in non_unique_params else param_name
|
248 |
-
cmd_argument2group_param[new_param_name] = [group_name, param_name]
|
249 |
-
group2param2cmdarg[group_name][param_name]=new_param_name
|
250 |
-
|
251 |
-
renamed_group[new_param_name] = param_info
|
252 |
-
renamed_config[group_name] = renamed_group
|
253 |
-
return renamed_config, cmd_argument2group_param, group2param2cmdarg
|
254 |
-
|
255 |
-
@staticmethod
|
256 |
-
def create_parser(config: dict) -> argparse.ArgumentParser:
|
257 |
-
"""
|
258 |
-
Create and configure an argparse parser based on the given configuration.
|
259 |
-
|
260 |
-
This method sets up a command-line argument parser with arguments defined in the configuration.
|
261 |
-
Each top-level key in the configuration represents a group of related arguments.
|
262 |
-
|
263 |
-
:param config: A dictionary where each key is a group name and each value is a dict
|
264 |
-
of parameters for that group. Each parameter's information should include
|
265 |
-
its type, default value, and help description.
|
266 |
-
:type config: dict
|
267 |
-
|
268 |
-
:return: Configured argparse.ArgumentParser instance with arguments added as specified
|
269 |
-
in the configuration.
|
270 |
-
:rtype: argparse.ArgumentParser
|
271 |
-
|
272 |
-
:raises ValueError: If an unknown or unsupported type is specified for a parameter.
|
273 |
-
"""
|
274 |
-
parser = argparse.ArgumentParser(description="Command-line parser for project settings")
|
275 |
-
# Mapping of type strings to Python types
|
276 |
-
type_mapping = {
|
277 |
-
'integer': int,
|
278 |
-
'int': int,
|
279 |
-
'float': float,
|
280 |
-
'string': str,
|
281 |
-
'str': str,
|
282 |
-
'bool': bool,
|
283 |
-
'boolean': bool,
|
284 |
-
'list': list
|
285 |
-
# Complex types like 'dict' and 'type' are intentionally excluded
|
286 |
-
}
|
287 |
-
|
288 |
-
# List of types to handle as strings
|
289 |
-
handle_as_string = ['dict', 'type', 'list']
|
290 |
-
excluded_parameters = ['vocabmap', 'np_tokentype', 'pretraining_dataset_data', 'optim']
|
291 |
-
|
292 |
-
|
293 |
-
for group_name, parameters in config.items():
|
294 |
-
group = parser.add_argument_group(group_name)
|
295 |
-
for param_name, param_info in parameters.items():
|
296 |
-
param_type_str = param_info['type']
|
297 |
-
description = param_info['description']
|
298 |
-
escaped_description = re.sub(r"([^%])%", r"\1%%", description)
|
299 |
-
if param_name in excluded_parameters:
|
300 |
-
continue
|
301 |
-
if param_type_str in handle_as_string:
|
302 |
-
# Handle these types as strings in argparse, conversion will be done later in the program
|
303 |
-
param_type = str
|
304 |
-
elif param_type_str not in type_mapping:
|
305 |
-
raise ValueError(f"Unknown or unsupported type '{param_type_str}' for parameter '{param_name}'")
|
306 |
-
else:
|
307 |
-
param_type = type_mapping[param_type_str]
|
308 |
-
|
309 |
-
#print(f'The current type is: {param_type}')
|
310 |
-
default_param = param_info['default']
|
311 |
-
description = param_info['description']
|
312 |
-
kwargs = {
|
313 |
-
'type': param_type,
|
314 |
-
'default': param_info['default'],
|
315 |
-
'help': escaped_description
|
316 |
-
} # Add constraints if they exist
|
317 |
-
"""
|
318 |
-
if 'constraints' in param_info:
|
319 |
-
constraints = param_info['constraints']
|
320 |
-
if 'min' in constraints:
|
321 |
-
kwargs['type'] = lambda x: eval(param_type_str)(x) if eval(param_type_str)(x) >= constraints['min'] else sys.exit(f"Value for {param_name} must be at least {constraints['min']}")
|
322 |
-
if 'max' in constraints:
|
323 |
-
kwargs['type'] = lambda x: eval(param_type_str)(x) if eval(param_type_str)(x) <= constraints['max'] else sys.exit(f"Value for {param_name} must be at most {constraints['max']}")
|
324 |
-
if 'options' in constraints:
|
325 |
-
kwargs['choices'] = constraints['options']
|
326 |
-
"""
|
327 |
-
# Add argument to the group
|
328 |
-
group.add_argument(f'--{param_name}', **kwargs)
|
329 |
-
#parser = add_hf_args_to_parser(parser)
|
330 |
-
|
331 |
-
return parser
|
332 |
-
|
333 |
-
|
334 |
-
|
335 |
-
class SeqConfig(BaseConfig):
|
336 |
-
"""Class to manage and validate sequence processing configurations."""
|
337 |
-
|
338 |
-
def __init__(self):
|
339 |
-
super().__init__()
|
340 |
-
self.default_seq_config_file = self._get_default_sequence_processing_config_file()
|
341 |
-
with open(self.default_seq_config_file, 'r') as file:
|
342 |
-
self.parameters = yaml.safe_load(file)
|
343 |
-
|
344 |
-
# Some postprocessing steps
|
345 |
-
self.parameters['tokenization']['shift']['constraints']['max'] = self.parameters['tokenization']['kmer']['default']-1
|
346 |
-
# Ha valaki update-li a k-mer paramter-t, akkor triggerelni kellene, hogy mi legyen.
|
347 |
-
|
348 |
-
self.get_and_set_segmentation_parameters()
|
349 |
-
self.get_and_set_tokenization_parameters()
|
350 |
-
self.get_and_set_computational_parameters()
|
351 |
-
|
352 |
-
def _get_default_sequence_processing_config_file(self) -> str:
|
353 |
-
"""
|
354 |
-
Retrieve the default sequence processing configuration file.
|
355 |
-
|
356 |
-
:return: Path to the configuration file.
|
357 |
-
:rtype: str
|
358 |
-
"""
|
359 |
-
current_path = pathlib.Path(__file__).parent
|
360 |
-
prokbert_seq_config_file = join(current_path, 'configs', 'sequence_processing.yaml')
|
361 |
-
self.current_path = current_path
|
362 |
-
|
363 |
-
try:
|
364 |
-
# Attempt to read the environment variable
|
365 |
-
prokbert_seq_config_file = os.environ['SEQ_CONFIG_FILE']
|
366 |
-
except KeyError:
|
367 |
-
# Handle the case when the environment variable is not found
|
368 |
-
pass
|
369 |
-
# print("SEQ_CONFIG_FILE environment variable has not been set. Using default value: {0}".format(prokbert_seq_config_file))
|
370 |
-
return prokbert_seq_config_file
|
371 |
-
|
372 |
-
|
373 |
-
def get_and_set_segmentation_parameters(self, parameters: dict = {}) -> dict:
|
374 |
-
"""
|
375 |
-
Retrieve and validate the provided parameters for segmentation.
|
376 |
-
|
377 |
-
:param parameters: A dictionary of parameters to be validated.
|
378 |
-
:type parameters: dict
|
379 |
-
:return: A dictionary of validated segmentation parameters.
|
380 |
-
:rtype: dict
|
381 |
-
:raises ValueError: If an invalid segmentation parameter is provided.
|
382 |
-
"""
|
383 |
-
segmentation_params = {k: self.get_parameter('segmentation', k) for k in self.parameters['segmentation']}
|
384 |
-
|
385 |
-
for param, param_value in parameters.items():
|
386 |
-
if param not in segmentation_params:
|
387 |
-
raise ValueError(f"The provided {param} is an INVALID segmentation parameter! The valid parameters are: {list(segmentation_params.keys())}")
|
388 |
-
self.validate('segmentation', param, param_value)
|
389 |
-
segmentation_params[param] = param_value
|
390 |
-
self.segmentation_params = segmentation_params
|
391 |
-
|
392 |
-
|
393 |
-
return segmentation_params
|
394 |
-
|
395 |
-
|
396 |
-
def get_and_set_tokenization_parameters(self, parameters: dict = {}) -> dict:
|
397 |
-
# Updating the other parameters if necesseary, i.e. if k-mer has-been changed, then the shift is updated and we run a parameter check at the end
|
398 |
-
|
399 |
-
tokenization_params = {k: self.get_parameter('tokenization', k) for k in self.parameters['tokenization']}
|
400 |
-
for param, param_value in parameters.items():
|
401 |
-
if param not in tokenization_params:
|
402 |
-
raise ValueError(f"The provided {param} is an INVALID tokenization parameter! The valid parameters are: {list(tokenization_params.keys())}")
|
403 |
-
self.validate('tokenization', param, param_value)
|
404 |
-
tokenization_params[param] = param_value
|
405 |
-
|
406 |
-
# Loading and check the vocab file. It is assumed that its ordered dictionary
|
407 |
-
vocabfile=tokenization_params['vocabfile']
|
408 |
-
act_kmer = tokenization_params['kmer']
|
409 |
-
if vocabfile=='auto':
|
410 |
-
vocabfile_path = join(self.current_path, 'data/prokbert_vocabs/', f'prokbert-base-dna{act_kmer}', 'vocab.txt')
|
411 |
-
tokenization_params['vocabfile'] = vocabfile_path
|
412 |
-
else:
|
413 |
-
vocabfile_path = vocabfile
|
414 |
-
with open(vocabfile_path) as vocabfile_in:
|
415 |
-
vocabmap = {line.strip(): i for i, line in enumerate(vocabfile_in)}
|
416 |
-
tokenization_params['vocabmap'] = vocabmap
|
417 |
-
|
418 |
-
# Loading the vocab
|
419 |
-
self.tokenization_params = tokenization_params
|
420 |
-
return tokenization_params
|
421 |
-
|
422 |
-
def get_and_set_computational_parameters(self, parameters: dict = {}) -> dict:
|
423 |
-
""" Reading and validating the computational paramters
|
424 |
-
"""
|
425 |
-
|
426 |
-
computational_params = {k: self.get_parameter('computation', k) for k in self.parameters['computation']}
|
427 |
-
core_count = cpu_count()
|
428 |
-
|
429 |
-
if computational_params['cpu_cores_for_segmentation'] == -1:
|
430 |
-
computational_params['cpu_cores_for_segmentation'] = core_count
|
431 |
-
|
432 |
-
if computational_params['cpu_cores_for_tokenization'] == -1:
|
433 |
-
computational_params['cpu_cores_for_tokenization'] = core_count
|
434 |
-
|
435 |
-
|
436 |
-
|
437 |
-
for param, param_value in parameters.items():
|
438 |
-
if param not in computational_params:
|
439 |
-
raise ValueError(f"The provided {param} is an INVALID computation parameter! The valid parameters are: {list(computational_params.keys())}")
|
440 |
-
self.validate('computation', param, param_value)
|
441 |
-
computational_params[param] = param_value
|
442 |
-
|
443 |
-
np_tokentype= SeqConfig.numpy_dtype_mapping[computational_params['numpy_token_integer_prec_byte']]
|
444 |
-
computational_params['np_tokentype'] = np_tokentype
|
445 |
-
self.computational_params = computational_params
|
446 |
-
return computational_params
|
447 |
-
|
448 |
-
|
449 |
-
def get_maximum_segment_length_from_token_count_from_params(self):
|
450 |
-
"""Calculating the maximum length of the segment from the token count """
|
451 |
-
max_token_counts = self.tokenization_params['token_limit']
|
452 |
-
shift = self.tokenization_params['shift']
|
453 |
-
kmer = self.tokenization_params['kmer']
|
454 |
-
return self.get_maximum_segment_length_from_token_count(max_token_counts, shift, kmer)
|
455 |
-
|
456 |
-
def get_maximum_token_count_from_max_length_from_params(self):
|
457 |
-
"""Calculating the maximum length of the segment from the token count """
|
458 |
-
|
459 |
-
|
460 |
-
max_segment_length = self.tokenization_params['max_segment_length']
|
461 |
-
shift = self.tokenization_params['shift']
|
462 |
-
kmer = self.tokenization_params['kmer']
|
463 |
-
max_token_count = self.get_maximum_token_count_from_max_length(max_segment_length, shift, kmer)
|
464 |
-
|
465 |
-
return max_token_count
|
466 |
-
|
467 |
-
def get_cmd_arg_parser(self) -> tuple[argparse.ArgumentParser, dict, dict]:
|
468 |
-
"""
|
469 |
-
Create and return a command-line argument parser for ProkBERT configurations, along with mappings
|
470 |
-
between command-line arguments and configuration parameters.
|
471 |
-
|
472 |
-
This method combines sequence configuration parameters with training configuration parameters
|
473 |
-
and sets up a command-line argument parser using these combined settings. It ensures that parameter
|
474 |
-
names are unique across different groups by renaming any non-unique parameters.
|
475 |
-
|
476 |
-
:return: A tuple containing:
|
477 |
-
- Configured argparse.ArgumentParser instance for handling ProkBERT configurations.
|
478 |
-
- A dictionary mapping new command-line arguments to their original group and parameter name.
|
479 |
-
- A dictionary mapping each group to a dict that maps the original parameter names
|
480 |
-
to the new command-line argument names.
|
481 |
-
:rtype: tuple[argparse.ArgumentParser, dict, dict]
|
482 |
-
|
483 |
-
Note: The method assumes that the configuration parameters for training and sequence configuration
|
484 |
-
are available within the class.
|
485 |
-
"""
|
486 |
-
combined_params = deepcopy(self.parameters)
|
487 |
-
combined_params['Sequence'] = {}
|
488 |
-
combined_params['Sequence']['fasta_file_dir'] = {'default': 'None',
|
489 |
-
'description' : 'Directory where the input fasta file are located for the pretraining',
|
490 |
-
'type': 'string'}
|
491 |
-
combined_params['Sequence']['out'] = {'default': 'pretrain.h5',
|
492 |
-
'description' : 'Output path',
|
493 |
-
'type': 'string'}
|
494 |
-
|
495 |
-
|
496 |
-
combined_params, cmd_argument2group_param, group2param2cmdarg = BaseConfig.rename_non_unique_parameters(combined_params)
|
497 |
-
|
498 |
-
parser = BaseConfig.create_parser(combined_params)
|
499 |
-
return parser,cmd_argument2group_param, group2param2cmdarg
|
500 |
-
|
501 |
-
|
502 |
-
@staticmethod
|
503 |
-
def get_maximum_segment_length_from_token_count(max_token_counts, shift, kmer):
|
504 |
-
"""Calcuates how long sequence can be covered
|
505 |
-
"""
|
506 |
-
|
507 |
-
max_segment_length = (max_token_counts-3)*shift + kmer
|
508 |
-
return max_segment_length
|
509 |
-
|
510 |
-
@staticmethod
|
511 |
-
def get_maximum_token_count_from_max_length(max_segment_length, shift, kmer):
|
512 |
-
"""Calcuates how long sequence can be covered
|
513 |
-
"""
|
514 |
-
max_token_count = int(np.ceil((max_segment_length - kmer)/shift+3))
|
515 |
-
return max_token_count
|
516 |
-
|
517 |
-
class ProkBERTConfig(BaseConfig):
|
518 |
-
"""Class to manage and validate pretraining configurations."""
|
519 |
-
|
520 |
-
torch_dtype_mapping = {1: torch.uint8,
|
521 |
-
2: torch.int16,
|
522 |
-
8: torch.int64,
|
523 |
-
4: torch.int32}
|
524 |
-
|
525 |
-
def __init__(self):
|
526 |
-
super().__init__()
|
527 |
-
|
528 |
-
self.default_pretrain_config_file = self._get_default_pretrain_config_file()
|
529 |
-
with open(self.default_pretrain_config_file, 'r') as file:
|
530 |
-
self.parameters = yaml.safe_load(file)
|
531 |
-
|
532 |
-
# Load and validate each parameter set
|
533 |
-
self.data_collator_params = self.get_set_parameters('data_collator')
|
534 |
-
self.model_params = self.get_set_parameters('model')
|
535 |
-
self.dataset_params = self.get_set_parameters('dataset')
|
536 |
-
self.pretraining_params = self.get_set_parameters('pretraining')
|
537 |
-
self.finetuning_params = self.get_set_parameters('finetuning')
|
538 |
-
# Getting the sequtils params as well
|
539 |
-
|
540 |
-
self.def_seq_config = SeqConfig()
|
541 |
-
self.segmentation_params = self.def_seq_config.get_and_set_segmentation_parameters(self.parameters['segmentation'])
|
542 |
-
self.tokenization_params = self.def_seq_config.get_and_set_tokenization_parameters(self.parameters['tokenization'])
|
543 |
-
self.computation_params = self.def_seq_config.get_and_set_computational_parameters(self.parameters['computation'])
|
544 |
-
|
545 |
-
self.default_torchtype = ProkBERTConfig.torch_dtype_mapping[self.computation_params['numpy_token_integer_prec_byte']]
|
546 |
-
|
547 |
-
hf_training_args = TrainingArguments("working_dir")
|
548 |
-
self.hf_training_args_dict = hf_training_args.to_dict()
|
549 |
-
|
550 |
-
|
551 |
-
def _get_default_pretrain_config_file(self) -> str:
|
552 |
-
"""
|
553 |
-
Retrieve the default pretraining configuration file.
|
554 |
-
|
555 |
-
:return: Path to the configuration file.
|
556 |
-
:rtype: str
|
557 |
-
"""
|
558 |
-
current_path = pathlib.Path(__file__).parent
|
559 |
-
pretrain_config_file = join(current_path, 'configs', 'pretraining.yaml')
|
560 |
-
|
561 |
-
try:
|
562 |
-
# Attempt to read the environment variable
|
563 |
-
pretrain_config_file = os.environ['PRETRAIN_CONFIG_FILE']
|
564 |
-
except KeyError:
|
565 |
-
# Handle the case when the environment variable is not found
|
566 |
-
pass
|
567 |
-
# print(f"PRETRAIN_CONFIG_FILE environment variable has not been set. Using default value: {pretrain_config_file}")
|
568 |
-
return pretrain_config_file
|
569 |
-
|
570 |
-
def get_set_parameters(self, parameter_class: str, parameters: dict = {}) -> dict:
|
571 |
-
"""
|
572 |
-
Retrieve and validate the provided parameters for a given parameter class.
|
573 |
-
|
574 |
-
:param parameter_class: The class/category of the parameter (e.g., 'data_collator').
|
575 |
-
:type parameter_class: str
|
576 |
-
:param parameters: A dictionary of parameters to be validated.
|
577 |
-
:type parameters: dict
|
578 |
-
:return: A dictionary of validated parameters.
|
579 |
-
:rtype: dict
|
580 |
-
:raises ValueError: If an invalid parameter is provided.
|
581 |
-
"""
|
582 |
-
class_params = {k: self.get_parameter(parameter_class, k) for k in self.parameters[parameter_class]}
|
583 |
-
|
584 |
-
|
585 |
-
# First validatiading the class parameters as well
|
586 |
-
for param, param_value in class_params.items():
|
587 |
-
|
588 |
-
self.validate(parameter_class, param, param_value)
|
589 |
-
|
590 |
-
|
591 |
-
for param, param_value in parameters.items():
|
592 |
-
if param not in class_params and (parameter_class!='pretraining'):
|
593 |
-
raise ValueError(f"The provided {param} is an INVALID {parameter_class} parameter! The valid parameters are: {list(class_params.keys())}")
|
594 |
-
else:
|
595 |
-
if parameter_class == 'pretraining' or parameter_class == 'finetuning' :
|
596 |
-
if param in self.hf_training_args_dict or param in class_params:
|
597 |
-
if param in class_params:
|
598 |
-
self.validate(parameter_class, param, param_value)
|
599 |
-
class_params[param] = param_value
|
600 |
-
else:
|
601 |
-
raise ValueError(f"The provided {param} is an INVALID {parameter_class} parameter! In addition is not a valid training argument.")
|
602 |
-
else:
|
603 |
-
self.validate(parameter_class, param, param_value)
|
604 |
-
class_params[param] = param_value
|
605 |
-
|
606 |
-
return class_params
|
607 |
-
|
608 |
-
def get_and_set_model_parameters(self, parameters: dict = {}) -> dict:
|
609 |
-
""" Setting the model parameters """
|
610 |
-
|
611 |
-
# Here we include the additional training arguments available for the trainer
|
612 |
-
|
613 |
-
self.model_params = self.get_set_parameters('model', parameters)
|
614 |
-
|
615 |
-
return self.model_params
|
616 |
-
|
617 |
-
def get_and_set_dataset_parameters(self, parameters: dict = {}) -> dict:
|
618 |
-
""" Setting the dataset parameters """
|
619 |
-
|
620 |
-
self.dataset_params = self.get_set_parameters('dataset', parameters)
|
621 |
-
|
622 |
-
return self.dataset_params
|
623 |
-
|
624 |
-
def get_and_set_pretraining_parameters(self, parameters: dict = {}) -> dict:
|
625 |
-
""" Setting the model parameters """
|
626 |
-
self.pretraining_params = self.get_set_parameters('pretraining', parameters)
|
627 |
-
|
628 |
-
return self.pretraining_params
|
629 |
-
|
630 |
-
|
631 |
-
def get_and_set_datacollator_parameters(self, parameters: dict = {}) -> dict:
|
632 |
-
""" Setting the model parameters """
|
633 |
-
self.data_collator_params = self.get_set_parameters('data_collator', parameters)
|
634 |
-
return self.data_collator_params
|
635 |
-
|
636 |
-
def get_and_set_segmentation_parameters(self, parameters: dict = {}) -> dict:
|
637 |
-
self.segmentation_params = self.def_seq_config.get_and_set_segmentation_parameters(parameters)
|
638 |
-
|
639 |
-
return self.segmentation_params
|
640 |
-
def get_and_set_tokenization_parameters(self, parameters: dict = {}) -> dict:
|
641 |
-
self.tokenization_params = self.def_seq_config.get_and_set_tokenization_parameters(parameters)
|
642 |
-
|
643 |
-
return self.tokenization_params
|
644 |
-
def get_and_set_computation_params(self, parameters: dict = {}) -> dict:
|
645 |
-
self.computation_params = self.def_seq_config.get_and_set_computational_parameters(parameters)
|
646 |
-
return self.computation_params
|
647 |
-
|
648 |
-
def get_and_set_finetuning_parameters(self, parameters: dict = {}) -> dict:
|
649 |
-
""" Setting the finetuning parameters """
|
650 |
-
|
651 |
-
# Here we include the additional training arguments available for the trainer
|
652 |
-
|
653 |
-
self.finetuning_params = self.get_set_parameters('finetuning', parameters)
|
654 |
-
|
655 |
-
return self.finetuning_params
|
656 |
-
|
657 |
-
|
658 |
-
def get_inference_parameters(self):
|
659 |
-
# Instantiate TrainingArguments to access default values
|
660 |
-
hf_defaults = TrainingArguments(output_dir="/tmp") # Dummy output_dir for initialization
|
661 |
-
|
662 |
-
return {
|
663 |
-
'inference': {
|
664 |
-
'fastain': {
|
665 |
-
'default': None,
|
666 |
-
'type': 'str',
|
667 |
-
'description': 'Path to the input data for inference.'
|
668 |
-
},
|
669 |
-
'out': {
|
670 |
-
'default': None,
|
671 |
-
'type': 'str',
|
672 |
-
'description': 'Output path for the inference results.'
|
673 |
-
},
|
674 |
-
'per_device_eval_batch_size': {
|
675 |
-
'default': hf_defaults.per_device_eval_batch_size,
|
676 |
-
'type': 'int',
|
677 |
-
'description': 'Batch size per device during evaluation.'
|
678 |
-
},
|
679 |
-
'ddp_backend': {
|
680 |
-
'default': hf_defaults.ddp_backend,
|
681 |
-
'type': 'str',
|
682 |
-
'description': 'The backend to use for distributed training.'
|
683 |
-
},
|
684 |
-
'dataloader_drop_last': {
|
685 |
-
'default': hf_defaults.dataloader_drop_last,
|
686 |
-
'type': 'bool',
|
687 |
-
'description': 'Drop the last incomplete batch if it is not divisible by the batch size.'
|
688 |
-
},
|
689 |
-
'torch_compile': {
|
690 |
-
'default': getattr(hf_defaults, 'torch_compile', False), # Fallback for compatibility
|
691 |
-
'type': 'bool',
|
692 |
-
'description': 'Whether to use TorchScript’s JIT compilation to accelerate training.'
|
693 |
-
},
|
694 |
-
'torch_compile_mode': {
|
695 |
-
'default': getattr(hf_defaults, 'torch_compile_mode', 'eager'), # Fallback for compatibility
|
696 |
-
'type': 'str',
|
697 |
-
'description': 'The JIT mode to use for compiling PyTorch operations.'
|
698 |
-
}
|
699 |
-
}
|
700 |
-
}
|
701 |
-
|
702 |
-
|
703 |
-
def get_cmd_arg_parser(self, keyset=[]) -> tuple[argparse.ArgumentParser, dict, dict]:
|
704 |
-
"""
|
705 |
-
Create and return a command-line argument parser for ProkBERT configurations, along with mappings
|
706 |
-
between command-line arguments and configuration parameters.
|
707 |
-
|
708 |
-
This method combines sequence configuration parameters with training configuration parameters
|
709 |
-
and sets up a command-line argument parser using these combined settings. It ensures that parameter
|
710 |
-
names are unique across different groups by renaming any non-unique parameters.
|
711 |
-
|
712 |
-
:return: A tuple containing:
|
713 |
-
- Configured argparse.ArgumentParser instance for handling ProkBERT configurations.
|
714 |
-
- A dictionary mapping new command-line arguments to their original group and parameter name.
|
715 |
-
- A dictionary mapping each group to a dict that maps the original parameter names
|
716 |
-
to the new command-line argument names.
|
717 |
-
:rtype: tuple[argparse.ArgumentParser, dict, dict]
|
718 |
-
|
719 |
-
Note: The method assumes that the configuration parameters for training and sequence configuration
|
720 |
-
are available within the class.
|
721 |
-
"""
|
722 |
-
if len(keyset) ==0:
|
723 |
-
trainin_conf_keysets = ['data_collator', 'model', 'dataset', 'pretraining', 'finetuning']
|
724 |
-
else:
|
725 |
-
trainin_conf_keysets = keyset
|
726 |
-
|
727 |
-
inference_params = self.get_inference_parameters()
|
728 |
-
seq_config = deepcopy(self.def_seq_config.parameters)
|
729 |
-
default_other_config = deepcopy(self.parameters)
|
730 |
-
combined_params = {}
|
731 |
-
for k,v in seq_config.items():
|
732 |
-
combined_params[k] = v
|
733 |
-
for k in trainin_conf_keysets:
|
734 |
-
combined_params[k] = default_other_config[k]
|
735 |
-
combined_params.update(inference_params)
|
736 |
-
combined_params, cmd_argument2group_param, group2param2cmdarg = BaseConfig.rename_non_unique_parameters(combined_params)
|
737 |
-
parser = BaseConfig.create_parser(combined_params)
|
738 |
-
|
739 |
-
return parser,cmd_argument2group_param, group2param2cmdarg
|
740 |
-
|
741 |
-
|
742 |
-
def get_user_provided_args(args, parser):
|
743 |
-
"""
|
744 |
-
Extract arguments provided by the user from the parsed arguments.
|
745 |
-
|
746 |
-
Args:
|
747 |
-
args (argparse.Namespace): Parsed command-line arguments.
|
748 |
-
parser (argparse.ArgumentParser): The argument parser instance.
|
749 |
-
|
750 |
-
Returns:
|
751 |
-
dict: A dictionary of user-provided arguments and their values.
|
752 |
-
"""
|
753 |
-
|
754 |
-
user_provided_args = {}
|
755 |
-
for action in parser._actions:
|
756 |
-
arg_name = action.dest
|
757 |
-
default_value = action.default
|
758 |
-
user_value = getattr(args, arg_name, None)
|
759 |
-
if user_value != default_value:
|
760 |
-
user_provided_args[arg_name] = user_value
|
761 |
-
|
762 |
-
return user_provided_args
|
763 |
-
|
764 |
-
|
765 |
-
|
766 |
-
|
767 |
-
|
768 |
-
|
769 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
general_utils.py
DELETED
@@ -1,309 +0,0 @@
|
|
1 |
-
# coding=utf-8
|
2 |
-
|
3 |
-
import pandas as pd
|
4 |
-
import os
|
5 |
-
import numpy as np
|
6 |
-
import subprocess
|
7 |
-
import shutil
|
8 |
-
""" Library for general utils, such as dataframe properties checking,
|
9 |
-
creating directories, checking files, etc.
|
10 |
-
"""
|
11 |
-
|
12 |
-
|
13 |
-
def check_expected_columns(df: pd.DataFrame, expected_columns: list) -> bool:
|
14 |
-
"""Checks if a DataFrame contains the expected columns.
|
15 |
-
|
16 |
-
:param df: The input DataFrame to be checked.
|
17 |
-
:type df: pd.DataFrame
|
18 |
-
:param expected_columns: A list of columns that are expected to be present in the DataFrame.
|
19 |
-
:type expected_columns: list
|
20 |
-
:param df: pd.DataFrame:
|
21 |
-
:param expected_columns: list:
|
22 |
-
:returns: True if all expected columns are present in the DataFrame, False otherwise.
|
23 |
-
:rtype: bool
|
24 |
-
:raises ValueError: If any of the expected columns are not present in the DataFrame.
|
25 |
-
|
26 |
-
Examples
|
27 |
-
--------
|
28 |
-
>>> df = pd.DataFrame({'A': [1, 2], 'B': [3, 4]})
|
29 |
-
>>> check_expected_columns(df, ['A', 'B'])
|
30 |
-
True
|
31 |
-
|
32 |
-
>>> check_expected_columns(df, ['A', 'C'])
|
33 |
-
ValueError: The following columns are missing: ['C']
|
34 |
-
"""
|
35 |
-
|
36 |
-
missing_columns = [col for col in expected_columns if col not in df.columns]
|
37 |
-
|
38 |
-
if missing_columns:
|
39 |
-
raise ValueError(f"The following columns are missing: {missing_columns}")
|
40 |
-
|
41 |
-
return True
|
42 |
-
|
43 |
-
|
44 |
-
def is_valid_primary_key(df: pd.DataFrame, column_name: str) -> bool:
|
45 |
-
"""Checks if a specified column in a DataFrame can serve as a valid primary key.
|
46 |
-
|
47 |
-
:param df: The input DataFrame to be checked.
|
48 |
-
:type df: pd.DataFrame
|
49 |
-
:param column_name: The name of the column to check.
|
50 |
-
:type column_name: str
|
51 |
-
:returns: True if the column can serve as a valid primary key, False otherwise.
|
52 |
-
:rtype: bool
|
53 |
-
:raises ValueError: If the specified column does not exist in the DataFrame.
|
54 |
-
|
55 |
-
Examples
|
56 |
-
--------
|
57 |
-
>>> df = pd.DataFrame({'A': [1, 2, 3], 'B': [4, 5, 6]})
|
58 |
-
>>> is_valid_primary_key(df, 'A')
|
59 |
-
True
|
60 |
-
|
61 |
-
>>> df = pd.DataFrame({'A': [1, 2, 2], 'B': [4, 5, 6]})
|
62 |
-
>>> is_valid_primary_key(df, 'A')
|
63 |
-
False
|
64 |
-
"""
|
65 |
-
|
66 |
-
if column_name not in df.columns:
|
67 |
-
raise ValueError(f"Column '{column_name}' does not exist in the DataFrame.")
|
68 |
-
|
69 |
-
# Check for NaN values
|
70 |
-
if df[column_name].isnull().any():
|
71 |
-
return False
|
72 |
-
|
73 |
-
# Check for unique values
|
74 |
-
if not df[column_name].is_unique:
|
75 |
-
return False
|
76 |
-
|
77 |
-
return True
|
78 |
-
|
79 |
-
def get_non_empty_files(start_path: str, extensions: tuple = ('.fasta', '.fna')) -> str:
|
80 |
-
"""Generator that yields non-empty files from a specified directory and its subdirectories based on the given extensions.
|
81 |
-
|
82 |
-
:param start_path: The path to the directory from which to start the search.
|
83 |
-
:type start_path: str
|
84 |
-
:param extensions: A tuple of file extensions to look for (default is ('.fasta', '.fna')).
|
85 |
-
The function also automatically checks for compressed versions with '.gz'.
|
86 |
-
:type extensions: tuple
|
87 |
-
:returns: Yields filenames that match the specified extensions and are non-empty.
|
88 |
-
:rtype: str
|
89 |
-
|
90 |
-
"""
|
91 |
-
|
92 |
-
for dirpath, _, filenames in os.walk(start_path):
|
93 |
-
for filename in filenames:
|
94 |
-
filepath = os.path.join(dirpath, filename)
|
95 |
-
if any(filename.endswith(ext) or filename.endswith(ext + '.gz') for ext in extensions) and os.path.getsize(filepath) > 0:
|
96 |
-
yield filename
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
def truncate_zero_columns(arr: np.ndarray) -> np.ndarray:
|
101 |
-
"""Truncate all trailing columns composed entirely of zeros in a given 2D numpy array.
|
102 |
-
|
103 |
-
:param arr: Input 2D numpy array.
|
104 |
-
:type arr: np.ndarray
|
105 |
-
:returns: A new array with trailing zero columns removed.
|
106 |
-
:rtype: np.ndarray
|
107 |
-
|
108 |
-
"""
|
109 |
-
|
110 |
-
# Iterate over columns from the end
|
111 |
-
for idx in range(arr.shape[1]-1, -1, -1):
|
112 |
-
if np.any(arr[:, idx]):
|
113 |
-
return arr[:, :(idx+1)]
|
114 |
-
return np.empty((arr.shape[0], 0))
|
115 |
-
|
116 |
-
|
117 |
-
import os
|
118 |
-
|
119 |
-
def create_directory_for_filepath(filepath: str) -> None:
|
120 |
-
"""Given a file path, creates the underlying directory structure if it doesn't already exist.
|
121 |
-
|
122 |
-
:param filepath: The path to the file for which the directory structure should be created.
|
123 |
-
:type filepath: str
|
124 |
-
:raises ValueError: If the provided path is empty or None.
|
125 |
-
:raises OSError: If there's an error creating the directory structure.
|
126 |
-
|
127 |
-
"""
|
128 |
-
|
129 |
-
if not filepath:
|
130 |
-
raise ValueError("The provided filepath is empty or None.")
|
131 |
-
|
132 |
-
directory = os.path.dirname(filepath)
|
133 |
-
|
134 |
-
if directory and not os.path.exists(directory):
|
135 |
-
try:
|
136 |
-
os.makedirs(directory)
|
137 |
-
print(f"Directory structure {directory} created successfully.")
|
138 |
-
except OSError as e:
|
139 |
-
raise OSError(f"Error creating directory structure {directory}. Error: {e}")
|
140 |
-
|
141 |
-
# Example usage:
|
142 |
-
# create_directory_for_filepath("/path/to/directory/that/might/not/exist/filename.txt")
|
143 |
-
|
144 |
-
def check_file_exists(file_path: str) -> bool:
|
145 |
-
"""Checks if the provided file path exists.
|
146 |
-
|
147 |
-
:param file_path: Path to the file.
|
148 |
-
:type file_path: str
|
149 |
-
:returns: True if the file exists, raises ValueError otherwise.
|
150 |
-
:rtype: bool
|
151 |
-
|
152 |
-
"""
|
153 |
-
if os.path.exists(file_path):
|
154 |
-
return True
|
155 |
-
else:
|
156 |
-
raise ValueError(f"The provided file path '{file_path}' does not exist.")
|
157 |
-
|
158 |
-
def count_gpus(method="clinfo"):
|
159 |
-
"""
|
160 |
-
Count the number of available GPUs using the specified method.
|
161 |
-
|
162 |
-
This function counts the number of NVIDIA and AMD GPUs using the chosen method. By default, it uses the 'clinfo'
|
163 |
-
method for AMD GPUs.
|
164 |
-
|
165 |
-
:param method: The method to use for GPU counting. Choose between 'clinfo' (default) and 'rocm'.
|
166 |
-
:type method: str, optional
|
167 |
-
|
168 |
-
:return: The total number of GPUs detected.
|
169 |
-
:rtype: int
|
170 |
-
|
171 |
-
:raises ValueError: If an unknown method is provided.
|
172 |
-
|
173 |
-
:raises Exception: If an error occurs while querying AMD GPUs using the specified method.
|
174 |
-
|
175 |
-
.. note::
|
176 |
-
- The 'clinfo' method queries AMD GPUs by running the 'clinfo' command.
|
177 |
-
- The 'rocm' method queries AMD GPUs by running 'rocm-smi --list' command.
|
178 |
-
|
179 |
-
"""
|
180 |
-
import torch
|
181 |
-
import subprocess
|
182 |
-
|
183 |
-
# Count NVIDIA GPUs
|
184 |
-
nvidia_gpu_count = torch.cuda.device_count()
|
185 |
-
|
186 |
-
# Count AMD GPUs
|
187 |
-
amd_gpu_count = 0
|
188 |
-
try:
|
189 |
-
if method == "clinfo":
|
190 |
-
clinfo_output = subprocess.check_output('clinfo').decode('utf-8')
|
191 |
-
amd_gpu_count = clinfo_output.lower().count('device type: gpu')
|
192 |
-
elif method == "rocm":
|
193 |
-
rocm_output = subprocess.check_output('rocm-smi --list', shell=True).decode('utf-8')
|
194 |
-
amd_gpu_count = len(rocm_output.strip().split('\n'))
|
195 |
-
else:
|
196 |
-
raise ValueError("Unknown method provided. Choose between 'clinfo' and 'rocm'.")
|
197 |
-
except Exception as e:
|
198 |
-
print(f"Error querying AMD GPUs using method '{method}': {e}")
|
199 |
-
|
200 |
-
total_gpus = nvidia_gpu_count + amd_gpu_count
|
201 |
-
|
202 |
-
return total_gpus
|
203 |
-
|
204 |
-
|
205 |
-
|
206 |
-
def create_hard_links(source_directory: str, target_directory: str, blacklist: list = []) -> None:
|
207 |
-
"""Creates hard links for all files from the source directory to the target directory.
|
208 |
-
|
209 |
-
:param source_directory: The directory containing the original files.
|
210 |
-
:type source_directory: str
|
211 |
-
:param target_directory: The directory where hard links will be created.
|
212 |
-
:type target_directory: str
|
213 |
-
:param blacklist: List of filenames to exclude from creating hard links.
|
214 |
-
:type blacklist: list
|
215 |
-
:returns: None
|
216 |
-
|
217 |
-
"""
|
218 |
-
|
219 |
-
# Ensure the provided directories exist
|
220 |
-
if not os.path.exists(source_directory):
|
221 |
-
raise ValueError(f"The source directory '{source_directory}' does not exist.")
|
222 |
-
if not os.path.exists(target_directory):
|
223 |
-
os.makedirs(target_directory)
|
224 |
-
|
225 |
-
# Iterate through the files in the source directory
|
226 |
-
for filename in os.listdir(source_directory):
|
227 |
-
source_file_path = os.path.join(source_directory, filename)
|
228 |
-
target_file_path = os.path.join(target_directory, filename)
|
229 |
-
|
230 |
-
# Check for files to skip
|
231 |
-
if (filename.startswith('.') or
|
232 |
-
filename.startswith('_') or
|
233 |
-
os.path.isdir(source_file_path) or
|
234 |
-
filename in blacklist):
|
235 |
-
continue
|
236 |
-
|
237 |
-
# Create a hard link
|
238 |
-
os.link(source_file_path, target_file_path)
|
239 |
-
|
240 |
-
return f"Hard links created in {target_directory} from {source_directory}."
|
241 |
-
|
242 |
-
# Example usage
|
243 |
-
# create_hard_links("/path/to/source_directory", "/path/to/target_directory", blacklist=["file_to_skip.txt"])
|
244 |
-
|
245 |
-
def create_selected_hard_links(source_directory: str, target_directory: str, filenames: list) -> None:
|
246 |
-
"""Creates hard links for the specified files from the source directory to the target directory.
|
247 |
-
|
248 |
-
:param source_directory: The directory containing the original files.
|
249 |
-
:type source_directory: str
|
250 |
-
:param target_directory: The directory where hard links will be created.
|
251 |
-
:type target_directory: str
|
252 |
-
:param filenames: List of filenames for which hard links should be created.
|
253 |
-
:type filenames: list
|
254 |
-
:returns: None
|
255 |
-
|
256 |
-
"""
|
257 |
-
|
258 |
-
# Ensure the provided directories exist
|
259 |
-
if not os.path.exists(source_directory):
|
260 |
-
raise ValueError(f"The source directory '{source_directory}' does not exist.")
|
261 |
-
if not os.path.exists(target_directory):
|
262 |
-
os.makedirs(target_directory)
|
263 |
-
|
264 |
-
# Iterate through the specified filenames
|
265 |
-
for filename in filenames:
|
266 |
-
source_file_path = os.path.join(source_directory, filename)
|
267 |
-
target_file_path = os.path.join(target_directory, filename)
|
268 |
-
|
269 |
-
# Ensure the file exists in the source directory
|
270 |
-
if not os.path.isfile(source_file_path):
|
271 |
-
print(f"Warning: {filename} does not exist in the source directory. Skipping.")
|
272 |
-
continue
|
273 |
-
|
274 |
-
# Create a hard link
|
275 |
-
try:
|
276 |
-
os.link(source_file_path, target_file_path)
|
277 |
-
except FileExistsError:
|
278 |
-
print(f'The target hard link {target_file_path} exist. Skipping...')
|
279 |
-
|
280 |
-
return f"Hard links for specified files created in {target_directory} from {source_directory}."
|
281 |
-
|
282 |
-
def remove_hidden_files(directory: str) -> None:
|
283 |
-
"""Removes all files recursively in a folder that start with '.' or '_'.
|
284 |
-
|
285 |
-
:param directory: The directory from which hidden files should be removed.
|
286 |
-
:type directory: str
|
287 |
-
:returns: None
|
288 |
-
|
289 |
-
"""
|
290 |
-
|
291 |
-
# Ensure the directory exists
|
292 |
-
if not os.path.exists(directory):
|
293 |
-
raise ValueError(f"The directory '{directory}' does not exist.")
|
294 |
-
|
295 |
-
# Use os.walk to iterate through all subdirectories and files
|
296 |
-
for dirpath, dirnames, filenames in os.walk(directory, topdown=False):
|
297 |
-
|
298 |
-
# Filter out directories starting with '.' or '_'
|
299 |
-
dirnames[:] = [d for d in dirnames if not d.startswith('.') and not d.startswith('_')]
|
300 |
-
|
301 |
-
# Remove files starting with '.' or '_'
|
302 |
-
for filename in filenames:
|
303 |
-
if filename.startswith('.') or filename.startswith('_'):
|
304 |
-
file_path = os.path.join(dirpath, filename)
|
305 |
-
os.remove(file_path)
|
306 |
-
print(f"Removed: {file_path}")
|
307 |
-
|
308 |
-
print(f"All hidden files removed from {directory}.")
|
309 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
sequtils.py
DELETED
@@ -1,980 +0,0 @@
|
|
1 |
-
|
2 |
-
import logging
|
3 |
-
|
4 |
-
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
|
5 |
-
# coding=utf-8
|
6 |
-
|
7 |
-
""" Library for sequence processing """
|
8 |
-
|
9 |
-
|
10 |
-
import os
|
11 |
-
import sys
|
12 |
-
import pandas as pd
|
13 |
-
from multiprocessing import Pool
|
14 |
-
import multiprocessing
|
15 |
-
from os.path import join, isfile, splitext
|
16 |
-
from os import listdir
|
17 |
-
import random
|
18 |
-
from Bio import SeqIO
|
19 |
-
import numpy as np
|
20 |
-
import math
|
21 |
-
import gzip
|
22 |
-
from mimetypes import guess_type
|
23 |
-
from functools import partial
|
24 |
-
import operator
|
25 |
-
import pathlib
|
26 |
-
#from typing import Dict, List, Type, Tuple
|
27 |
-
from itertools import product
|
28 |
-
from typing import List, Union, Dict, Any, Optional, Tuple, Type, Set
|
29 |
-
from .general_utils import *
|
30 |
-
from Bio.Seq import Seq
|
31 |
-
from Bio.SeqRecord import SeqRecord
|
32 |
-
from scipy.ndimage import convolve1d
|
33 |
-
import h5py
|
34 |
-
|
35 |
-
def load_contigs(
|
36 |
-
fasta_files_list: Union[List[str], str],
|
37 |
-
adding_reverse_complement: bool = True,
|
38 |
-
IsAddHeader: bool = False,
|
39 |
-
AsDataFrame: bool = False,
|
40 |
-
to_uppercase: bool = False,
|
41 |
-
is_add_sequence_id: bool = False
|
42 |
-
) -> Union[List[Union[str, List[str]]], pd.DataFrame]:
|
43 |
-
"""
|
44 |
-
Loads contigs from a list of FASTA files.
|
45 |
-
|
46 |
-
:param fasta_files_list: List of paths to FASTA files or a single file path. Compressed (gz) FASTA files are accepted.
|
47 |
-
:type fasta_files_list: Union[List[str], str]
|
48 |
-
:param adding_reverse_complement: If True, adds the reverse complement of each sequence. Defaults to True.
|
49 |
-
:type adding_reverse_complement: bool
|
50 |
-
:param IsAddHeader: If True, includes the FASTA ID and description in the output. Defaults to False.
|
51 |
-
:type IsAddHeader: bool
|
52 |
-
:param AsDataFrame: If True, returns the sequences as a pandas DataFrame. Defaults to False.
|
53 |
-
:type AsDataFrame: bool
|
54 |
-
:param to_uppercase: If True, converts sequences to uppercase. Defaults to False.
|
55 |
-
:type to_uppercase: bool
|
56 |
-
:param is_add_sequence_id: If True, adds a unique integer sequence ID to each sequence. Defaults to False.
|
57 |
-
:type is_add_sequence_id: bool
|
58 |
-
:return: The loaded sequences. Each sequence is represented as a string if IsAddHeader is False, or as a list
|
59 |
-
[sequence_id, fasta_id, description, source_file, sequence, orientation] if IsAddHeader is True and is_add_sequence_id is True.
|
60 |
-
If AsDataFrame is True, the sequences are returned as a DataFrame.
|
61 |
-
:rtype: Union[List[Union[str, List[str]]], pd.DataFrame]
|
62 |
-
|
63 |
-
Example:
|
64 |
-
>>> fasta_files = ['path/to/file1.fasta', 'path/to/file2.fasta.gz']
|
65 |
-
>>> load_contigs(fasta_files, adding_reverse_complement=False, IsAddHeader=True, AsDataFrame=True, to_uppercase=True, is_add_sequence_id=True)
|
66 |
-
# Returns a DataFrame with the sequences from the specified FASTA files, all in uppercase, with unique sequence IDs.
|
67 |
-
"""
|
68 |
-
|
69 |
-
logging.info('Loading sequence data into memory!')
|
70 |
-
if isinstance(fasta_files_list, str):
|
71 |
-
logging.info('Since the fasta_files_list is a string, not a list, we convert it to a list.')
|
72 |
-
fasta_files_list = [fasta_files_list]
|
73 |
-
|
74 |
-
sequences = []
|
75 |
-
sequence_id = 0
|
76 |
-
df_cols = ['sequence_id', 'fasta_id', 'description', 'source_file', 'sequence', 'orientation'] if (IsAddHeader and is_add_sequence_id) else ['fasta_id', 'description', 'source_file', 'sequence', 'orientation'] if IsAddHeader else ['sequence']
|
77 |
-
for act_assembly in fasta_files_list:
|
78 |
-
# Determine the file encoding based on the file extension
|
79 |
-
encoding = guess_type(act_assembly)[1]
|
80 |
-
_open = partial(gzip.open, mode='rt') if encoding == 'gzip' else open
|
81 |
-
with _open(act_assembly) as f_assembly:
|
82 |
-
# Parse the fasta file
|
83 |
-
contigs = list(SeqIO.parse(f_assembly, "fasta"))
|
84 |
-
for contig in contigs:
|
85 |
-
act_seq = str(contig.seq)[:] if not to_uppercase else str(contig.seq).upper()[:]
|
86 |
-
act_header = str(contig.id)
|
87 |
-
act_description = str(contig.description)
|
88 |
-
if adding_reverse_complement:
|
89 |
-
# Compute the reverse complement of the sequence
|
90 |
-
act_reverse_complement = str(contig.seq.reverse_complement()) if not to_uppercase else str(contig.seq.reverse_complement()).upper()
|
91 |
-
|
92 |
-
if IsAddHeader:
|
93 |
-
# Include sequence ID (if applicable), fasta ID, description, source file, sequence, and orientation in the output
|
94 |
-
entry = [sequence_id] if is_add_sequence_id else []
|
95 |
-
entry.extend([act_header, act_description, act_assembly, act_seq, 'forward'])
|
96 |
-
sequences.append(entry)
|
97 |
-
if adding_reverse_complement:
|
98 |
-
entry = [sequence_id + 1] if is_add_sequence_id else []
|
99 |
-
entry.extend([act_header, act_description, act_assembly, act_reverse_complement, 'reverse'])
|
100 |
-
sequences.append(entry)
|
101 |
-
if is_add_sequence_id:
|
102 |
-
sequence_id += 2
|
103 |
-
else:
|
104 |
-
sequence_id+=1
|
105 |
-
else:
|
106 |
-
# Only include the sequence in the output
|
107 |
-
sequences.append(act_seq)
|
108 |
-
if adding_reverse_complement:
|
109 |
-
sequences.append(act_reverse_complement)
|
110 |
-
|
111 |
-
if AsDataFrame:
|
112 |
-
# Convert the sequences to a DataFrame
|
113 |
-
sequences = pd.DataFrame(sequences, columns=df_cols)
|
114 |
-
return sequences
|
115 |
-
|
116 |
-
|
117 |
-
def segment_sequence_contiguous(
|
118 |
-
sequence: str,
|
119 |
-
params: Dict[str, Any],
|
120 |
-
sequence_id: Optional[Any] = np.nan
|
121 |
-
) -> List[Dict[str, Any]]:
|
122 |
-
"""
|
123 |
-
Creates end-to-end, disjoint segments of a sequence without overlaps.
|
124 |
-
|
125 |
-
Segments smaller than the predefined minimum length will be discarded.
|
126 |
-
This function returns a list of segments along with their positions in the original sequence.
|
127 |
-
|
128 |
-
:param sequence: The input nucleotide sequence to be segmented.
|
129 |
-
:type sequence: str
|
130 |
-
:param params: Dictionary containing the segmentation parameters. Must include 'min_length' and 'max_length' keys
|
131 |
-
specifying the minimum and maximum lengths of the segments, respectively. Can contain other parameters.
|
132 |
-
:type params: Dict[str, Any]
|
133 |
-
:param sequence_id: An identifier for the sequence, optional. Defaults to NaN.
|
134 |
-
:type sequence_id: Optional[Any]
|
135 |
-
:return: A list of dictionaries, each representing a segment. Each dictionary contains the segment's sequence,
|
136 |
-
start position, end position, and sequence ID.
|
137 |
-
:rtype: List[Dict[str, Any]]
|
138 |
-
|
139 |
-
Example:
|
140 |
-
>>> params = {'min_length': 0, 'max_length': 100}
|
141 |
-
>>> segment_sequence_contiguous('ATCGATCGA', params)
|
142 |
-
[{'segment': 'ATCGATCGA', 'segment_start': 0, 'segment_end': 9, 'sequence_id': np.nan}]
|
143 |
-
"""
|
144 |
-
|
145 |
-
# Extract segmentation parameters
|
146 |
-
min_segment_len = params['min_length']
|
147 |
-
max_segment_len = params['max_length']
|
148 |
-
|
149 |
-
# Ensure the sequence is treated as a string
|
150 |
-
if isinstance(sequence, str):
|
151 |
-
act_seq = sequence
|
152 |
-
L = len(sequence)
|
153 |
-
|
154 |
-
segments = []
|
155 |
-
for i in range(0, L, max_segment_len):
|
156 |
-
act_start_pos = i
|
157 |
-
act_end_pos = min(i + max_segment_len, L)
|
158 |
-
act_segment = sequence[act_start_pos:act_end_pos]
|
159 |
-
|
160 |
-
|
161 |
-
|
162 |
-
# Add segment to the list if it's longer than the minimum length
|
163 |
-
if len(act_segment) >= min_segment_len:
|
164 |
-
new_record = {
|
165 |
-
'segment': act_segment,
|
166 |
-
'segment_start': act_start_pos,
|
167 |
-
'segment_end': act_end_pos,
|
168 |
-
'sequence_id': sequence_id
|
169 |
-
}
|
170 |
-
segments.append(new_record)
|
171 |
-
|
172 |
-
return segments
|
173 |
-
|
174 |
-
|
175 |
-
|
176 |
-
def segment_sequences_random(
|
177 |
-
sequences: Union[pd.DataFrame, List[str]],
|
178 |
-
params: Dict[str, Union[int, float, str, Dict, List, Tuple]]
|
179 |
-
) -> List[Dict[str, Union[int, str]]]:
|
180 |
-
"""
|
181 |
-
Randomly segments the input sequences.
|
182 |
-
|
183 |
-
This function accepts either a list of sequences or a DataFrame containing sequences.
|
184 |
-
If a DataFrame is provided, it's assumed to have preprocessed sequences with "sequence" and "sequence_id" columns,
|
185 |
-
where "sequence_id" is a valid primary key. The function returns a list of dictionaries,
|
186 |
-
each containing details of a segment including its sequence, start position, end position,
|
187 |
-
associated sequence ID, and a segment ID (not generated in this function).
|
188 |
-
|
189 |
-
:param sequences: A DataFrame containing sequences with "sequence" and "sequence_id" columns or a list of sequences.
|
190 |
-
:type sequences: Union[pd.DataFrame, List[str]]
|
191 |
-
:param params: Dictionary containing segmentation parameters such as 'coverage', 'min_length', and 'max_length'.
|
192 |
-
:type params: Dict[str, Union[int, float, str, Dict, List, Tuple]]
|
193 |
-
:return: A list of dictionaries with each containing details of a segment.
|
194 |
-
:rtype: List[Dict[str, Union[int, str]]]
|
195 |
-
|
196 |
-
Notes:
|
197 |
-
- The actual number of segments may differ from the expected number due to random sampling and sequences
|
198 |
-
being shorter than the specified segment size.
|
199 |
-
- Segment IDs are not generated by this function.
|
200 |
-
"""
|
201 |
-
|
202 |
-
# Calculate sequence lengths and cumulative sum of lengths
|
203 |
-
sequences['seq_lengths'] = sequences.apply(lambda x: len(x['sequence']), axis=1)
|
204 |
-
sequences['lenght_cum_sum'] = sequences['seq_lengths'].cumsum()
|
205 |
-
Lseqs = sum(sequences['seq_lengths'])
|
206 |
-
|
207 |
-
# Calculate the number of segments to sample based on expected coverage.
|
208 |
-
# Note: The actual number might be biased if many sequences are "short" compared to the segment sizes.
|
209 |
-
N_segments = int(np.ceil(params['coverage'] * Lseqs / params['max_length']))
|
210 |
-
logging.info(f'Sampling {N_segments} segments from {len(sequences)} sequences.')
|
211 |
-
|
212 |
-
# Generate random starting coordinates for segments
|
213 |
-
start_coords = list(np.sort(np.int64(np.random.uniform(0, sequences['lenght_cum_sum'].max(), N_segments))))
|
214 |
-
segmentdb = []
|
215 |
-
|
216 |
-
for sid, act_sampling_coord in enumerate(start_coords):
|
217 |
-
|
218 |
-
diff = act_sampling_coord - sequences['lenght_cum_sum']
|
219 |
-
|
220 |
-
# Find the sequence in which the current segment starts
|
221 |
-
for i in range(len(sequences['lenght_cum_sum'])):
|
222 |
-
if diff[i] < 0:
|
223 |
-
break
|
224 |
-
|
225 |
-
act_sequence_id = sequences['sequence_id'].iloc[i]
|
226 |
-
rel_coord = act_sampling_coord - sequences['lenght_cum_sum'].iloc[i] + sequences['seq_lengths'].iloc[i]
|
227 |
-
|
228 |
-
segment_end = min(rel_coord + params['max_length'], sequences['seq_lengths'].iloc[i])
|
229 |
-
|
230 |
-
# Skip the segment if it's shorter than the minimum segment length
|
231 |
-
if segment_end - rel_coord < params['min_length']:
|
232 |
-
pred_seqgment = sequences['sequence'].iloc[i][rel_coord:segment_end]
|
233 |
-
minimum_len = params['min_length']
|
234 |
-
logging.info(f'Too short segment, skip! Sampled segment: {pred_seqgment}, Segment end coordinate: {segment_end}, relative coordinate: {rel_coord}, minimum length is: {minimum_len}')
|
235 |
-
continue
|
236 |
-
|
237 |
-
new_segment = sequences['sequence'].iloc[i][rel_coord:segment_end]
|
238 |
-
new_record = {
|
239 |
-
'sequence_id': act_sequence_id,
|
240 |
-
'segment_start': rel_coord,
|
241 |
-
'segment_end': segment_end,
|
242 |
-
'segment': new_segment,
|
243 |
-
'segment_id': str(sid)
|
244 |
-
}
|
245 |
-
|
246 |
-
segmentdb.append(new_record)
|
247 |
-
|
248 |
-
return segmentdb
|
249 |
-
|
250 |
-
def segment_sequences(
|
251 |
-
sequences: Union[List[str], pd.DataFrame],
|
252 |
-
params: Dict[str, Union[int, float, str, ]],
|
253 |
-
AsDataFrame: bool = False
|
254 |
-
) -> Union[List[str], pd.DataFrame]:
|
255 |
-
"""
|
256 |
-
Segments sequences based on the provided parameters.
|
257 |
-
|
258 |
-
This function assumes that the sequence is quality controlled and preprocessed, i.e., it is a valid nucleotide sequence.
|
259 |
-
If sequences are provided as a DataFrame, then it is assumed that there is a "sequence_id" and
|
260 |
-
a "sequence" attribute. The "sequence_id" should be a valid primary key.
|
261 |
-
If the output is requested as a DataFrame, then the IDs are added as well.
|
262 |
-
|
263 |
-
:param sequences: A list of sequences or a DataFrame containing sequences.
|
264 |
-
If a DataFrame, it must have "sequence_id" and "sequence" attributes.
|
265 |
-
:type sequences: Union[List[str], pd.DataFrame]
|
266 |
-
:param params: Dictionary containing the segmentation parameters.
|
267 |
-
- 'type' (str): The type of segmentation ('contiguous' or 'random').
|
268 |
-
- 'min_length' (int): Minimum length of a segment.
|
269 |
-
- 'max_length' (int): Maximum length of a segment.
|
270 |
-
- 'coverage' (float): Coverage percentage for random segmentation.
|
271 |
-
:type params: Dict[str, Union[int, float, str, Dict[str, int], List[int], Tuple[int, int]]]
|
272 |
-
:param AsDataFrame: If True, the output will be a DataFrame. If False, it will be a list. Defaults to False.
|
273 |
-
:type AsDataFrame: bool
|
274 |
-
:return: List of segmented sequences or a DataFrame with segmented sequences and their corresponding information based on the `AsDataFrame` parameter.
|
275 |
-
:rtype: Union[List[str], pd.DataFrame]
|
276 |
-
:raises ValueError: If the provided sequences DataFrame does not have the required attributes.
|
277 |
-
:raises ValueError: If the "sequence_id" column is not a valid primary key.
|
278 |
-
|
279 |
-
Examples:
|
280 |
-
>>> segment_sequences(['AATCAATTTTATTT', 'AGCCGATTCAATTGCATTATTT'], {'type': 'contiguous', 'min_length': 1, 'max_length': 1000, 'coverage': 1.0})
|
281 |
-
"""
|
282 |
-
|
283 |
-
segmentation_type = params['type']
|
284 |
-
|
285 |
-
# Checking for primary key and sequence attribute???
|
286 |
-
expected_attributes = ['sequence_id', 'sequence']
|
287 |
-
return_cols = ['segment_id', 'sequence_id', 'segment_start', 'segment_end', 'segment']
|
288 |
-
|
289 |
-
if isinstance(sequences, list):
|
290 |
-
logging.info('Sequences is a list, therefore ignoring ids and tracking information. ')
|
291 |
-
IsSequenceId = None
|
292 |
-
IsSeqList = True
|
293 |
-
elif isinstance(sequences, pd.DataFrame):
|
294 |
-
#logging.info('Sequences is a list, therefore adding tracking information.')
|
295 |
-
logging.info('Checking input DataFrame!')
|
296 |
-
check_expected_columns(sequences, expected_attributes)
|
297 |
-
logging.info('Checking input sequence_id is valid primary key in the DataFrame')
|
298 |
-
is_valid_primary_key(sequences, 'sequence_id')
|
299 |
-
IsSequenceId = True
|
300 |
-
IsSeqList=False
|
301 |
-
|
302 |
-
segments = []
|
303 |
-
if segmentation_type == 'contiguous':
|
304 |
-
if IsSeqList:
|
305 |
-
if IsSequenceId:
|
306 |
-
for act_seq_id, seq in enumerate(sequences):
|
307 |
-
act_segments = segment_sequence_contiguous(seq, params, act_seq_id)
|
308 |
-
segments.extend(act_segments)
|
309 |
-
else:
|
310 |
-
for seq in sequences:
|
311 |
-
act_segments = segment_sequence_contiguous(seq, params)
|
312 |
-
segments.extend(act_segments)
|
313 |
-
else:
|
314 |
-
for _, rec in sequences.iterrows():
|
315 |
-
act_seq = rec['sequence']
|
316 |
-
act_seq_id = rec['sequence_id']
|
317 |
-
act_segments = segment_sequence_contiguous(act_seq, params, act_seq_id)
|
318 |
-
segments.extend(act_segments)
|
319 |
-
|
320 |
-
elif segmentation_type == 'random':
|
321 |
-
if IsSeqList:
|
322 |
-
seqeunce_df = pd.DataFrame(sequences,
|
323 |
-
columns = ['sequence'])
|
324 |
-
seqeunce_df['sequence_id'] = list(range(len(sequences)))
|
325 |
-
segments = segment_sequences_random(seqeunce_df, params)
|
326 |
-
|
327 |
-
else:
|
328 |
-
segments = segment_sequences_random(sequences, params)
|
329 |
-
if AsDataFrame:
|
330 |
-
#logging.info('Creating a DataFrame from the segments. ')
|
331 |
-
segment_db = pd.DataFrame(segments)
|
332 |
-
segment_ids = list(range(len(segment_db)))
|
333 |
-
segment_db['segment_id'] = segment_ids
|
334 |
-
segment_db = segment_db[return_cols]
|
335 |
-
|
336 |
-
else:
|
337 |
-
segment_db = [seg['segment'] for seg in segments]
|
338 |
-
return segment_db
|
339 |
-
|
340 |
-
def lca_kmer_tokenize_segment(segment: str, offset: int, params: Dict[str, Dict[str, int] | int | float]):
|
341 |
-
# calculate the tokenization for one offset value
|
342 |
-
shift = params['shift']
|
343 |
-
max_segment_length = params['max_segment_length']
|
344 |
-
max_unknown_token_proportion = params['max_unknown_token_proportion']
|
345 |
-
kmer = params['kmer']
|
346 |
-
token_limit = params['token_limit']
|
347 |
-
vocabmap = params['vocabmap']
|
348 |
-
add_special_token = params['add_special_token']
|
349 |
-
if len(segment) > max_segment_length:
|
350 |
-
raise(ValueError(f'The segment is longer {len(segment)} then the maximum allowed segment length ({max_segment_length}). '))
|
351 |
-
|
352 |
-
kmers = [segment[i:i + kmer] for i in range(offset, len(segment) - kmer + 1, shift)]
|
353 |
-
|
354 |
-
return kmers
|
355 |
-
|
356 |
-
|
357 |
-
|
358 |
-
|
359 |
-
|
360 |
-
def lca_tokenize_segment(segment: str, params: Dict[str, Dict[str, int] | int | float]) -> Tuple[List[List[int]], List[List[str]]]:
|
361 |
-
"""
|
362 |
-
Tokenizes a single segment using Local Context Aware (LCA) tokenization.
|
363 |
-
The segment is first split into k-mers with specified shifts and then tokenized into token vectors.
|
364 |
-
|
365 |
-
:param segment: The input nucleotide sequence segment to be tokenized.
|
366 |
-
:type segment: str
|
367 |
-
:param params: Dictionary containing the tokenization parameters.
|
368 |
-
- 'shift' (int): The k-mer shift parameter.
|
369 |
-
- 'max_segment_length' (int): Maximum allowable segment length.
|
370 |
-
- 'max_unknown_token_proportion' (float): Maximum allowable proportion of unknown tokens in a segment.
|
371 |
-
- 'kmer' (int): Size of the k-mer.
|
372 |
-
- 'token_limit' (int): Maximum number of tokens allowed in the tokenized output.
|
373 |
-
- 'vocabmap' (dict[str, int]): Dictionary mapping k-mers to their respective token values.
|
374 |
-
:type params: dict
|
375 |
-
:returns: A tuple containing:
|
376 |
-
- list[list[int]]: List of tokenized segments (each segment as a list of integers).
|
377 |
-
- list[list[str]]: List of k-merized segments with different shifts (each segment as a list of strings).
|
378 |
-
:rtype: Tuple[List[List[int]], List[List[str]]]
|
379 |
-
:raises ValueError: If the segment length exceeds the `max_segment_length`.
|
380 |
-
|
381 |
-
Examples:
|
382 |
-
>>> vocabmap_example = {"[CLS]": 2, "[SEP]": 3, "[UNK]": 0, "TCTTT": 4, "CTTTG": 5, "TTTGC": 6, "TTGCT": 7}
|
383 |
-
>>> segment_example = 'TCTTTGCTAAG'
|
384 |
-
>>> params_example = {'shift': 1, 'max_segment_length': 512, 'max_unknown_token_proportion': 0.2, 'kmer': 5, 'token_limit': 10, 'vocabmap': vocabmap_example}
|
385 |
-
>>> lca_tokenize_segment(segment_example, params_example)
|
386 |
-
([[2, 4, 5, 6, 7, 3]], [['TCTTT', 'CTTTG', 'TTTGC', 'TTGCT']])
|
387 |
-
"""
|
388 |
-
|
389 |
-
|
390 |
-
#logging.info('Tokenizing a segment')
|
391 |
-
shift = params['shift']
|
392 |
-
max_segment_length = params['max_segment_length']
|
393 |
-
max_unknown_token_proportion = params['max_unknown_token_proportion']
|
394 |
-
kmer = params['kmer']
|
395 |
-
token_limit = params['token_limit']
|
396 |
-
vocabmap = params['vocabmap']
|
397 |
-
add_special_token = params['add_special_token']
|
398 |
-
if len(segment) > max_segment_length:
|
399 |
-
raise(ValueError(f'The segment is longer {len(segment)} then the maximum allowed segment length ({max_segment_length}). '))
|
400 |
-
|
401 |
-
kmers_offset = []
|
402 |
-
# For every pssoble offset and window we should get a k-mer vector.
|
403 |
-
# If the segmen is too short or non-existent, then we might have a problem. So, please ensure the segment
|
404 |
-
for offset in range(shift):
|
405 |
-
kmers = [segment[i:i + kmer] for i in range(offset, len(segment) - kmer + 1, shift)]
|
406 |
-
kmers_offset.append(kmers)
|
407 |
-
# Mapping the k-mers into numbers
|
408 |
-
tokenized_segments = tokenize_kmerized_segment_list(kmers_offset, vocabmap, token_limit, max_unknown_token_proportion, add_special_token)
|
409 |
-
return tokenized_segments, kmers_offset
|
410 |
-
|
411 |
-
|
412 |
-
|
413 |
-
def tokenize_kmerized_segment_list(kmerized_segments: List[List[str]],
|
414 |
-
vocabmap: Dict[str, int],
|
415 |
-
token_limit: int,
|
416 |
-
max_unknown_token_proportion: float,
|
417 |
-
add_special_tokens: bool = True) -> List[List[int]]:
|
418 |
-
"""Tokenizes or vectorizes a list of k-merized segments into a list of token vectors. If the expected number of
|
419 |
-
tokens in a segment exceeds the maximum allowed tokens (`token_limit`), the function raises an error. For segments
|
420 |
-
where unknown k-mers exceed the proportion set by `max_unknown_token_proportion`, the output is a special token
|
421 |
-
sequence indicating an empty sentence.
|
422 |
-
|
423 |
-
:param kmerized_segments: List containing k-merized segments.
|
424 |
-
:type kmerized_segments: List[List[str]]
|
425 |
-
:param vocabmap: Dictionary that maps k-mers to their respective token values.
|
426 |
-
:type vocabmap: Dict[str, int]
|
427 |
-
:param token_limit: Maximum number of tokens allowed in the tokenized output.
|
428 |
-
:type token_limit: int
|
429 |
-
:param max_unknown_token_proportion: Maximum allowable proportion of unknown tokens in a segment.
|
430 |
-
:type max_unknown_token_proportion: float
|
431 |
-
:param add_special_tokens: Whether to add special tokens (`[CLS]` and `[SEP]`) to the tokenized segments.
|
432 |
-
:type add_special_tokens: bool, optional (default=True)
|
433 |
-
:returns: List containing tokenized segments.
|
434 |
-
:rtype: List[List[int]]
|
435 |
-
:raises ValueError: If the expected number of tokens in a segment exceeds `token_limit`.
|
436 |
-
|
437 |
-
Examples
|
438 |
-
--------
|
439 |
-
|
440 |
-
>>> vocabmap_example = {"[CLS]": 2, "[SEP]": 3, "[UNK]": 0, "TCTTTG": 4, "CTTTGC": 5, "TTTGCT": 6, "TTGCTA": 7}
|
441 |
-
>>> kmerized_segment_example = [['TCTTTG', 'CTTTGC', 'TTTGCT', 'TTGCTA']]
|
442 |
-
>>> tokenize_kmerized_segment_list(kmerized_segment_example, vocabmap_example, 10, 0.2)
|
443 |
-
[[2, 4, 5, 6, 7, 3]]
|
444 |
-
"""
|
445 |
-
|
446 |
-
tokenized_segments = []
|
447 |
-
if add_special_tokens:
|
448 |
-
empty_sentence = [2, 3]
|
449 |
-
else:
|
450 |
-
empty_sentence = []
|
451 |
-
|
452 |
-
for act_kmer_list in kmerized_segments:
|
453 |
-
if add_special_tokens:
|
454 |
-
tokenized_kmerized_segment = [vocabmap['[CLS]']]
|
455 |
-
else:
|
456 |
-
tokenized_kmerized_segment = []
|
457 |
-
unkcount=0
|
458 |
-
L_kmerized_segment = len(act_kmer_list)
|
459 |
-
unkw_tsh_count = int(L_kmerized_segment*max_unknown_token_proportion)
|
460 |
-
if len(act_kmer_list)+2 > token_limit:
|
461 |
-
raise(ValueError(f'The expected number of tokens in the segment ({L_kmerized_segment+2}) is larger, then the maximum allowed number of tokens = ({token_limit}). '))
|
462 |
-
if L_kmerized_segment == 0:
|
463 |
-
logging.info('Its and empty sentence')
|
464 |
-
tokenized_kmerized_segment = empty_sentence
|
465 |
-
tokenized_segments.append(empty_sentence)
|
466 |
-
continue
|
467 |
-
for kmer in act_kmer_list:
|
468 |
-
try:
|
469 |
-
tokenized_kmerized_segment.append(vocabmap[kmer.upper()])
|
470 |
-
except KeyError:
|
471 |
-
tokenized_kmerized_segment.append(vocabmap['[UNK]'])
|
472 |
-
unkcount+=1
|
473 |
-
if unkcount > unkw_tsh_count:
|
474 |
-
tokenized_segments.append(empty_sentence)
|
475 |
-
continue
|
476 |
-
if add_special_tokens:
|
477 |
-
tokenized_kmerized_segment.append(vocabmap['[SEP]'])
|
478 |
-
tokenized_segments.append(tokenized_kmerized_segment)
|
479 |
-
|
480 |
-
return tokenized_segments
|
481 |
-
|
482 |
-
def process_batch_tokenize_segments_with_ids(
|
483 |
-
segments: List[str],
|
484 |
-
segment_ids: List[Any],
|
485 |
-
tokenization_params: Dict[str, Any],
|
486 |
-
np_token_type: type = np.uint16
|
487 |
-
) -> Dict[Any, List[np.ndarray]]:
|
488 |
-
"""
|
489 |
-
Tokenizes a batch of segments and associates them with their provided IDs.
|
490 |
-
|
491 |
-
This function generates vector representations for a collection of segments, assuming the segments
|
492 |
-
have undergone quality control. The result is a dictionary where the keys are segment IDs, and the values
|
493 |
-
are lists of potential vector representations for the segment, with each list element corresponding to
|
494 |
-
a specific shift.
|
495 |
-
|
496 |
-
The vector representations are converted to numpy arrays. The output is not a 2D rectangular array but
|
497 |
-
a dictionary mapping each segment ID to its tokenized representations.
|
498 |
-
|
499 |
-
:param segments: A list of preprocessed and validated segments.
|
500 |
-
:type segments: List[str]
|
501 |
-
:param segment_ids: A list of segment IDs corresponding to each segment in `segments`.
|
502 |
-
:type segment_ids: List[Any]
|
503 |
-
:param tokenization_params: A dictionary containing tokenization parameters.
|
504 |
-
:type tokenization_params: Dict[str, Any]
|
505 |
-
:param np_token_type: Numpy data type for the tokenized segments. Defaults to np.uint16.
|
506 |
-
:type np_token_type: type, optional
|
507 |
-
:return: A dictionary with segment IDs as keys and lists of numpy arrays representing tokenized segments as values.
|
508 |
-
:rtype: Dict[Any, List[np.ndarray]]
|
509 |
-
|
510 |
-
Example:
|
511 |
-
>>> segments = ['ACTG', 'TGCA']
|
512 |
-
>>> segment_ids = [1, 2]
|
513 |
-
>>> tokenization_params = {'max_segment_length': 50, ...}
|
514 |
-
>>> tokenized_segments = process_batch_tokenize_segments_with_ids(
|
515 |
-
segments, segment_ids, tokenization_params
|
516 |
-
)
|
517 |
-
"""
|
518 |
-
tokenized_segments_with_ids = {}
|
519 |
-
for i, segment in enumerate(segments):
|
520 |
-
act_id = segment_ids[i]
|
521 |
-
tokenized_segments_with_ids[act_id] = []
|
522 |
-
max_segment_length = tokenization_params['max_segment_length']
|
523 |
-
if len(segment) > max_segment_length:
|
524 |
-
raise ValueError(f'The segment is longer ({len(segment)}) than the maximum allowed segment length ({max_segment_length}).')
|
525 |
-
|
526 |
-
tokenized_segment, _ = lca_tokenize_segment(segment, tokenization_params)
|
527 |
-
tokenized_segment = [np.array(act_segment, dtype=np_token_type) for act_segment in tokenized_segment]
|
528 |
-
tokenized_segments_with_ids[act_id] = tokenized_segment
|
529 |
-
return tokenized_segments_with_ids
|
530 |
-
|
531 |
-
def batch_tokenize_segments_with_ids(
|
532 |
-
segment_data: Union[Tuple[List[str], List[Any]], pd.DataFrame],
|
533 |
-
tokenization_params: Dict[str, Any],
|
534 |
-
num_cores: int = 1,
|
535 |
-
batch_size: int = 10000,
|
536 |
-
np_token_type: type = np.uint16
|
537 |
-
) -> Dict[Any, List[np.ndarray]]:
|
538 |
-
"""
|
539 |
-
Parallel tokenization of segments with associated IDs.
|
540 |
-
|
541 |
-
This function splits the input data into batches and uses multiprocessing to tokenize
|
542 |
-
the segments in parallel. It supports both list/tuple inputs and pandas DataFrames.
|
543 |
-
|
544 |
-
:param segment_data: Either a tuple/list containing two elements (segments, segment_ids),
|
545 |
-
or a pandas DataFrame with 'segment' and 'segment_id' columns.
|
546 |
-
:type segment_data: Union[Tuple[List[str], List[Any]], pd.DataFrame]
|
547 |
-
:param tokenization_params: Dictionary containing tokenization parameters.
|
548 |
-
:type tokenization_params: Dict[str, Any]
|
549 |
-
:param num_cores: Number of CPU cores to use for parallel processing. Defaults to 1.
|
550 |
-
:type num_cores: int, optional
|
551 |
-
:param batch_size: Number of segments to process in each batch. Defaults to 10,000.
|
552 |
-
:type batch_size: int, optional
|
553 |
-
:param np_token_type: Numpy data type for the tokenized segments. Defaults to np.uint16.
|
554 |
-
:type np_token_type: type, optional
|
555 |
-
:return: A dictionary where keys are segment IDs and values are lists of numpy arrays representing tokenized segments.
|
556 |
-
:rtype: Dict[Any, List[np.ndarray]]
|
557 |
-
:raises ValueError: If the input data is neither a tuple/list nor a pandas DataFrame.
|
558 |
-
|
559 |
-
Example:
|
560 |
-
>>> segments = ['ACTG', 'TGCA']
|
561 |
-
>>> segment_ids = [1, 2]
|
562 |
-
>>> tokenization_params = {'max_segment_length': 50, ...}
|
563 |
-
>>> tokenized_data = batch_tokenize_segments_with_ids(
|
564 |
-
(segments, segment_ids),
|
565 |
-
tokenization_params,
|
566 |
-
num_cores=4,
|
567 |
-
batch_size=1000
|
568 |
-
)
|
569 |
-
"""
|
570 |
-
if isinstance(segment_data, tuple) or isinstance(segment_data, list):
|
571 |
-
segments = segment_data[0]
|
572 |
-
segment_ids = segment_data[1]
|
573 |
-
elif isinstance(segment_data, pd.DataFrame):
|
574 |
-
segments = list(segment_data['segment'])
|
575 |
-
segment_ids = list(segment_data['segment_id'])
|
576 |
-
else:
|
577 |
-
raise ValueError(f'The input should be either pandas DataFrame or a tuple instead of {type(segment_data)}')
|
578 |
-
|
579 |
-
Ndata = len(segments)
|
580 |
-
batch_intervals = [(i, min(i + batch_size, Ndata)) for i in range(0, Ndata, batch_size)]
|
581 |
-
params = [
|
582 |
-
(segments[interval[0]:interval[1]],
|
583 |
-
segment_ids[interval[0]:interval[1]],
|
584 |
-
tokenization_params,
|
585 |
-
np_token_type)
|
586 |
-
for interval in batch_intervals
|
587 |
-
]
|
588 |
-
with Pool(processes=num_cores) as pool:
|
589 |
-
result_list = pool.starmap(process_batch_tokenize_segments_with_ids, params)
|
590 |
-
|
591 |
-
tokenized_sets = {}
|
592 |
-
for d in result_list:
|
593 |
-
tokenized_sets.update(d)
|
594 |
-
|
595 |
-
return tokenized_sets
|
596 |
-
|
597 |
-
|
598 |
-
def get_rectangular_array_from_tokenized_dataset(tokenized_segments_data: Dict[int, List[np.ndarray]], shift: int, max_token_count: int, truncate_zeros: bool = True, randomize: bool = True, numpy_dtype: Type = np.uint16) -> Tuple[np.ndarray, pd.DataFrame]:
|
599 |
-
"""Create a rectangular numpy array that can be used as input to a Language Model (LM) from tokenized segment data.
|
600 |
-
|
601 |
-
:param tokenized_segments_data: A dictionary where keys are segment ids and values are lists of possible LCA tokenized vectors.
|
602 |
-
:type tokenized_segments_data: Dict[int, List[np.ndarray]]
|
603 |
-
|
604 |
-
:param shift: Number of LCA offsets.
|
605 |
-
:type shift: int
|
606 |
-
|
607 |
-
:param max_token_count: Maximum allowed token count in the output numpy array.
|
608 |
-
:type max_token_count: int
|
609 |
-
|
610 |
-
:param truncate_zeros: If True, truncate columns from the end of the numpy array that only contain zeros. (default=True)
|
611 |
-
:type truncate_zeros: bool, optional
|
612 |
-
|
613 |
-
:param randomize: If True, randomize the order of the rows in the output numpy array. (default=True)
|
614 |
-
:type randomize: bool, optional
|
615 |
-
|
616 |
-
:param numpy_dtype: Data type of the values in the output numpy array. (default=np.uint16)
|
617 |
-
:type numpy_dtype: Type, optional
|
618 |
-
|
619 |
-
:returns: A rectangular numpy array suitable for input to an LM.
|
620 |
-
:rtype: np.ndarray
|
621 |
-
|
622 |
-
:returns: A dataframe that describes which row in the numpy array corresponds to which segment and its LCA offset.
|
623 |
-
Columns are: ['torch_id', 'segment_id', 'offset']
|
624 |
-
:rtype: pd.DataFrame
|
625 |
-
|
626 |
-
"""
|
627 |
-
|
628 |
-
|
629 |
-
expected_length = len(tokenized_segments_data)*shift
|
630 |
-
X=np.full((expected_length,max_token_count),0, dtype=numpy_dtype)
|
631 |
-
torch_db = []
|
632 |
-
torch_id = 0
|
633 |
-
for segment_id, tokenized_vectors in tokenized_segments_data.items():
|
634 |
-
for offset in range(shift):
|
635 |
-
segment_vector = tokenized_vectors[offset]
|
636 |
-
X[torch_id,0:segment_vector.shape[0]] = segment_vector
|
637 |
-
torch_db.append([torch_id, segment_id, offset])
|
638 |
-
torch_id+=1
|
639 |
-
torch_tokenized_segment_db = pd.DataFrame(torch_db,
|
640 |
-
columns = ['torch_id', 'segment_id', 'offset'])
|
641 |
-
|
642 |
-
if randomize:
|
643 |
-
logging.info('Doing randomization!')
|
644 |
-
perm = np.random.permutation(expected_length)
|
645 |
-
X = X[perm,:]
|
646 |
-
torch_tokenized_segment_db.rename({'torch_id': 'original_torch_id'}, axis=1, inplace=True)
|
647 |
-
torch_tokenized_segment_db = torch_tokenized_segment_db.iloc[perm,:].reset_index().drop('index', axis=1).reset_index().rename({'index' : 'torch_id'}, axis=1)
|
648 |
-
|
649 |
-
if truncate_zeros:
|
650 |
-
logging.info('Tuncating all zeros column')
|
651 |
-
X = truncate_zero_columns(X)
|
652 |
-
return X, torch_tokenized_segment_db
|
653 |
-
|
654 |
-
|
655 |
-
def pretty_print_overlapping_sequence(segment, segment_kmers, tokenizer_params):
|
656 |
-
"""
|
657 |
-
Format the sequence for pretty printing with overlapping k-mers.
|
658 |
-
|
659 |
-
:param segment: DNA sequence.
|
660 |
-
:type segment: str
|
661 |
-
|
662 |
-
:param segment_kmers: List of k-mers in the segment.
|
663 |
-
:type segment_kmers: list
|
664 |
-
|
665 |
-
:param tokenizer_params: Dictionary containing tokenization parameters.
|
666 |
-
:type tokenizer_params: dict
|
667 |
-
|
668 |
-
:return: List of formatted strings representing the sequence with overlapping k-mers.
|
669 |
-
:rtype: list
|
670 |
-
"""
|
671 |
-
|
672 |
-
shift = tokenizer_params['shift']
|
673 |
-
k = tokenizer_params['kmer']
|
674 |
-
sep_c = 2
|
675 |
-
lines = []
|
676 |
-
base_offset = len(str( int((k+3)/shift))) + 3
|
677 |
-
first_line = ' '*base_offset + segment
|
678 |
-
lines.append(first_line)
|
679 |
-
nr_lines = int(np.ceil((k+sep_c)/shift))
|
680 |
-
logging.info('Nr. line to cover the seq: {0}'.format(nr_lines))
|
681 |
-
|
682 |
-
for line_id in range(nr_lines):
|
683 |
-
|
684 |
-
line_mers = [k_mer for j, k_mer in enumerate(segment_kmers) if j%nr_lines== line_id]
|
685 |
-
act_line = str(line_id) + '. ' + ' '*(line_id*shift) + (' '*(sep_c)).join(line_mers)
|
686 |
-
lines.append(act_line)
|
687 |
-
lines = '\n'.join(lines)
|
688 |
-
return lines
|
689 |
-
|
690 |
-
|
691 |
-
def generate_kmers(abc: Set[str], k: int) -> List[str]:
|
692 |
-
"""
|
693 |
-
Generates all possible k-mers from a given alphabet.
|
694 |
-
|
695 |
-
:param abc: The alphabet.
|
696 |
-
:type abc: Set[str]
|
697 |
-
:param k: Length of the k-mers.
|
698 |
-
:type k: int
|
699 |
-
:return: List of all possible k-mers.
|
700 |
-
:rtype: List[str]
|
701 |
-
"""
|
702 |
-
return [''.join(p) for p in product(abc, repeat=k)]
|
703 |
-
|
704 |
-
def save_to_hdf(X: np.ndarray, hdf_file_path: str, database: pd.DataFrame = None, compression: bool = False, pd_chunksize: int = 10_000_000) -> None:
|
705 |
-
"""Save a numpy array and an optional pandas DataFrame to an HDF5 file.
|
706 |
-
|
707 |
-
:param X: 2D numpy array to be saved.
|
708 |
-
:type X: np.ndarray
|
709 |
-
:param hdf_file_path: Path to the HDF5 file.
|
710 |
-
:type hdf_file_path: str
|
711 |
-
:param database: Pandas DataFrame to be saved. Defaults to None.
|
712 |
-
:type database: pd.DataFrame
|
713 |
-
:param compression: Whether to apply compression. Defaults to False.
|
714 |
-
:type compression: bool
|
715 |
-
:param pd_chunksize: Number of rows per chunk for saving the DataFrame. Defaults to 10,000,000.
|
716 |
-
:type pd_chunksize: int
|
717 |
-
:raises ValueError: If the provided numpy array is not 2D.
|
718 |
-
:raises OSError: If there's an error creating the directory structure or removing an existing HDF5 file.
|
719 |
-
Example:
|
720 |
-
|
721 |
-
>>> import numpy as np
|
722 |
-
>>> import pandas as pd
|
723 |
-
>>> array = np.random.random((100, 100))
|
724 |
-
>>> df = pd.DataFrame({'A': range(1, 101), 'B': range(101, 201)})
|
725 |
-
>>> save_to_hdf(array, "sample.hdf5", database=df, compression=True)
|
726 |
-
"""
|
727 |
-
|
728 |
-
# Check if X is a 2D numpy array
|
729 |
-
if len(X.shape) != 2:
|
730 |
-
raise ValueError("The provided numpy array is not 2D.")
|
731 |
-
|
732 |
-
# If HDF5 file exists, attempt to delete it
|
733 |
-
if os.path.exists(hdf_file_path):
|
734 |
-
try:
|
735 |
-
os.remove(hdf_file_path)
|
736 |
-
logging.info(f"Existing HDF5 file {hdf_file_path} removed successfully.")
|
737 |
-
except Exception as e:
|
738 |
-
raise OSError(f"Error removing existing HDF5 file {hdf_file_path}. Error: {e}")
|
739 |
-
|
740 |
-
# Create directory structure for HDF5 file
|
741 |
-
create_directory_for_filepath(hdf_file_path)
|
742 |
-
|
743 |
-
# Save the numpy array to HDF5
|
744 |
-
with h5py.File(hdf_file_path, 'w') as hdf:
|
745 |
-
try:
|
746 |
-
grp = hdf.create_group("training_data")
|
747 |
-
except ValueError:
|
748 |
-
del hdf['training_data']
|
749 |
-
|
750 |
-
if compression:
|
751 |
-
grp.create_dataset("X", data=X, compression="lzf", chunks=True)
|
752 |
-
else:
|
753 |
-
grp.create_dataset("X", data=X, chunks=True)
|
754 |
-
|
755 |
-
logging.info(f"Numpy array saved to {hdf_file_path} successfully.")
|
756 |
-
|
757 |
-
# Save the pandas DataFrame to HDF5, if provided
|
758 |
-
if database is not None:
|
759 |
-
logging.info("Adding database into the HDF5 file!")
|
760 |
-
num_chunks = int(np.ceil(len(database) / pd_chunksize))
|
761 |
-
logging.info(f'Number of chunks: {num_chunks}')
|
762 |
-
chunk_grouping = np.arange(len(database)) // pd_chunksize
|
763 |
-
chunkseqs = database.groupby(chunk_grouping)
|
764 |
-
for i, (_, chunk) in enumerate(chunkseqs):
|
765 |
-
logging.info(f'Writing database chunk {i} into {hdf_file_path}')
|
766 |
-
if compression:
|
767 |
-
chunk.to_hdf(hdf_file_path, f'database_{i}', format='table', data_columns=True, mode='a', complib='lzo')
|
768 |
-
else:
|
769 |
-
chunk.to_hdf(hdf_file_path, f'database_{i}', format='table', data_columns=True, mode='a')
|
770 |
-
|
771 |
-
logging.info('Database addition finished!')
|
772 |
-
|
773 |
-
|
774 |
-
|
775 |
-
def dataframe_to_seqrecords(
|
776 |
-
df: pd.DataFrame,
|
777 |
-
fastaidcol: str = 'test_fastaid',
|
778 |
-
sequencecol: str = 'sequence'
|
779 |
-
) -> List[SeqRecord]:
|
780 |
-
"""
|
781 |
-
Convert a DataFrame with sequence information into a list of SeqRecord objects.
|
782 |
-
|
783 |
-
:param df: DataFrame containing at least two columns: one for sequence IDs and one for sequences.
|
784 |
-
:type df: pd.DataFrame
|
785 |
-
:param fastaidcol: Name of the column in `df` that contains sequence IDs. Defaults to 'test_fastaid'.
|
786 |
-
:type fastaidcol: str, optional
|
787 |
-
:param sequencecol: Name of the column in `df` that contains nucleotide sequences. Defaults to 'sequence'.
|
788 |
-
:type sequencecol: str, optional
|
789 |
-
:return: A list of SeqRecord objects constructed from the DataFrame.
|
790 |
-
:rtype: List[SeqRecord]
|
791 |
-
|
792 |
-
Example:
|
793 |
-
>>> import pandas as pd
|
794 |
-
>>> data = {'test_fastaid': ['seq1', 'seq2'], 'sequence': ['ATCG', 'GGTA']}
|
795 |
-
>>> df = pd.DataFrame(data)
|
796 |
-
>>> seq_records = dataframe_to_seqrecords(df)
|
797 |
-
>>> seq_records[0].id
|
798 |
-
'seq1'
|
799 |
-
"""
|
800 |
-
seq_records = []
|
801 |
-
for _, row in df.iterrows():
|
802 |
-
seq = Seq(row[sequencecol])
|
803 |
-
record = SeqRecord(seq, id=str(row[fastaidcol]), description="")
|
804 |
-
seq_records.append(record)
|
805 |
-
return seq_records
|
806 |
-
|
807 |
-
|
808 |
-
def write_seqrecords_to_fasta(
|
809 |
-
seq_records: List[SeqRecord],
|
810 |
-
file_name: str
|
811 |
-
) -> None:
|
812 |
-
"""
|
813 |
-
Write a list of SeqRecord objects to a FASTA file.
|
814 |
-
|
815 |
-
:param seq_records: List of SeqRecord objects to be written to file.
|
816 |
-
:type seq_records: List[SeqRecord]
|
817 |
-
:param file_name: Name or path of the file to write the FASTA records.
|
818 |
-
:type file_name: str
|
819 |
-
:return: None
|
820 |
-
:rtype: None
|
821 |
-
|
822 |
-
Example:
|
823 |
-
>>> from Bio.Seq import Seq
|
824 |
-
>>> from Bio.SeqRecord import SeqRecord
|
825 |
-
>>> seq_records = [SeqRecord(Seq('ATCG'), id='seq1'), SeqRecord(Seq('GGTA'), id='seq2')]
|
826 |
-
>>> write_seqrecords_to_fasta(seq_records, 'output.fasta')
|
827 |
-
"""
|
828 |
-
SeqIO.write(seq_records, file_name, "fasta")
|
829 |
-
|
830 |
-
|
831 |
-
def dump_records_to_files(
|
832 |
-
seq_records: List[SeqRecord],
|
833 |
-
folder_path: str
|
834 |
-
) -> None:
|
835 |
-
"""
|
836 |
-
Write each SeqRecord to a separate FASTA file in the specified folder.
|
837 |
-
|
838 |
-
:param seq_records: List of SeqRecord objects to be written individually.
|
839 |
-
:type seq_records: List[SeqRecord]
|
840 |
-
:param folder_path: Path to the folder where the files should be saved.
|
841 |
-
The folder will be created if it does not exist.
|
842 |
-
:type folder_path: str
|
843 |
-
:return: None
|
844 |
-
:rtype: None
|
845 |
-
|
846 |
-
Example:
|
847 |
-
>>> from Bio.Seq import Seq
|
848 |
-
>>> from Bio.SeqRecord import SeqRecord
|
849 |
-
>>> seq_records = [SeqRecord(Seq('ATCG'), id='seq1'), SeqRecord(Seq('GGTA'), id='seq2')]
|
850 |
-
>>> dump_records_to_files(seq_records, 'sequences_folder')
|
851 |
-
"""
|
852 |
-
# Ensure the folder exists
|
853 |
-
os.makedirs(folder_path, exist_ok=True)
|
854 |
-
|
855 |
-
for record in seq_records:
|
856 |
-
file_path = os.path.join(folder_path, f"{record.id}.fasta")
|
857 |
-
SeqIO.write(record, file_path, "fasta")
|
858 |
-
|
859 |
-
|
860 |
-
def split_seqrecords_to_fasta_chunks(
|
861 |
-
seq_records: List[SeqRecord],
|
862 |
-
output_folder: str,
|
863 |
-
chunk_size_mb: int = 10
|
864 |
-
) -> None:
|
865 |
-
"""
|
866 |
-
Splits a list of SeqRecord objects into multiple FASTA files, each less than a specified size in MB.
|
867 |
-
|
868 |
-
:param seq_records: List of SeqRecord objects to be split into chunks.
|
869 |
-
:type seq_records: List[SeqRecord]
|
870 |
-
:param output_folder: The output folder where the FASTA files will be saved.
|
871 |
-
:type output_folder: str
|
872 |
-
:param chunk_size_mb: Maximum size of each FASTA file in megabytes. Defaults to 10 MB.
|
873 |
-
:type chunk_size_mb: int, optional
|
874 |
-
:return: None
|
875 |
-
:rtype: None
|
876 |
-
|
877 |
-
Example:
|
878 |
-
>>> seq_records = [...] # A list of SeqRecord objects
|
879 |
-
>>> split_seqrecords_to_fasta_chunks(seq_records, 'output_chunks', chunk_size_mb=5)
|
880 |
-
|
881 |
-
Notes:
|
882 |
-
- The last chunk may be smaller than the specified `chunk_size_mb`.
|
883 |
-
- The function approximates the size of each record for chunking.
|
884 |
-
"""
|
885 |
-
# Ensure output folder exists
|
886 |
-
os.makedirs(output_folder, exist_ok=True)
|
887 |
-
|
888 |
-
current_chunk = []
|
889 |
-
current_chunk_size = 0 # in bytes
|
890 |
-
chunk_id = 1 # Identifier for chunks/files
|
891 |
-
for record in seq_records:
|
892 |
-
# Approximate size of the record in bytes
|
893 |
-
record_size = len(str(record.seq)) + len(record.id) + 2 # Adding buffer for '>' and '\n'
|
894 |
-
|
895 |
-
# Check if adding this record exceeds the chunk size
|
896 |
-
if current_chunk_size + record_size > chunk_size_mb * 1024 * 1024:
|
897 |
-
file_path = os.path.join(output_folder, f"chunk_{chunk_id}.fasta")
|
898 |
-
SeqIO.write(current_chunk, file_path, "fasta")
|
899 |
-
current_chunk = []
|
900 |
-
current_chunk_size = 0
|
901 |
-
chunk_id += 1
|
902 |
-
|
903 |
-
current_chunk.append(record)
|
904 |
-
current_chunk_size += record_size
|
905 |
-
|
906 |
-
# Write any remaining records to the last chunk
|
907 |
-
if current_chunk:
|
908 |
-
file_path = os.path.join(output_folder, f"chunk_{chunk_id}.fasta")
|
909 |
-
SeqIO.write(current_chunk, file_path, "fasta")
|
910 |
-
|
911 |
-
|
912 |
-
def filter_short_sequences(
|
913 |
-
seq_records: List[SeqRecord],
|
914 |
-
length_threshold: int
|
915 |
-
) -> List[SeqRecord]:
|
916 |
-
"""
|
917 |
-
Filters out SeqRecord objects with sequences shorter than a specified threshold.
|
918 |
-
|
919 |
-
:param seq_records: List of SeqRecord objects.
|
920 |
-
:type seq_records: List[SeqRecord]
|
921 |
-
:param length_threshold: The minimum length of sequences to be retained.
|
922 |
-
:type length_threshold: int
|
923 |
-
:return: A list of SeqRecord objects that meet or exceed the length threshold.
|
924 |
-
:rtype: List[SeqRecord]
|
925 |
-
|
926 |
-
Example:
|
927 |
-
>>> from Bio.Seq import Seq
|
928 |
-
>>> from Bio.SeqRecord import SeqRecord
|
929 |
-
>>> records = [
|
930 |
-
... SeqRecord(Seq('ATCG'), id='seq1'),
|
931 |
-
... SeqRecord(Seq('AT'), id='seq2')
|
932 |
-
... ]
|
933 |
-
>>> filtered_records = filter_short_sequences(records, 3)
|
934 |
-
>>> len(filtered_records)
|
935 |
-
1
|
936 |
-
>>> filtered_records[0].id
|
937 |
-
'seq1'
|
938 |
-
"""
|
939 |
-
filtered_records = [record for record in seq_records if len(record.seq) >= length_threshold]
|
940 |
-
return filtered_records
|
941 |
-
|
942 |
-
|
943 |
-
|
944 |
-
def get_token_counts_for_segment(Lseg, kmer, shift, offset):
|
945 |
-
nr_tokens = int((Lseg -kmer)/shift + 1)
|
946 |
-
return nr_tokens
|
947 |
-
|
948 |
-
def get_seq_coordinates(token_pos, kmer, shift, offset):
|
949 |
-
seq_start = int(token_pos*shift + offset)
|
950 |
-
seq_end = int(token_pos*shift+kmer + offset)
|
951 |
-
return seq_start, seq_end
|
952 |
-
|
953 |
-
def get_token_coordinates(seq_pos, kmer, shift, offset, Lseg):
|
954 |
-
|
955 |
-
nrtokens = get_token_counts_for_segment(Lseg, kmer, shift, offset)
|
956 |
-
|
957 |
-
token_pos_end = int((seq_pos+offset - kmer) / shift)
|
958 |
-
token_pos_start = int((seq_pos + offset) / shift)
|
959 |
-
|
960 |
-
if token_pos_end<0:
|
961 |
-
token_pos_end=0
|
962 |
-
if token_pos_start >= nrtokens:
|
963 |
-
token_pos_start = nrtokens-1
|
964 |
-
|
965 |
-
return token_pos_start, token_pos_end
|
966 |
-
|
967 |
-
def sliding_window_average(arr, window_size=6):
|
968 |
-
# Create a window for averaging
|
969 |
-
window = np.ones(window_size) / window_size
|
970 |
-
# Use 'valid' mode to slide the window over the array without padding
|
971 |
-
result = np.convolve(arr, window, mode='valid')
|
972 |
-
return result
|
973 |
-
|
974 |
-
def convolve_expression_array(expression_array, window_size=6, step=2):
|
975 |
-
# Define the averaging window
|
976 |
-
window = np.ones(window_size) / window_size
|
977 |
-
# Apply convolution along each column (axis=0)
|
978 |
-
convolved_array = convolve1d(expression_array, window, axis=1, mode='reflect')
|
979 |
-
# Downsample by step size
|
980 |
-
return convolved_array[:, ::step]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
tokenizer.py
DELETED
@@ -1,363 +0,0 @@
|
|
1 |
-
import collections
|
2 |
-
import os
|
3 |
-
import json
|
4 |
-
from copy import deepcopy
|
5 |
-
from typing import List, Optional, Tuple, Dict
|
6 |
-
from transformers import PreTrainedTokenizer
|
7 |
-
from transformers.utils.hub import cached_file, hf_hub_url
|
8 |
-
|
9 |
-
from .config_utils import SeqConfig
|
10 |
-
from .sequtils import generate_kmers, lca_kmer_tokenize_segment
|
11 |
-
|
12 |
-
# Define the names of the vocabulary files
|
13 |
-
VOCAB_FILES_NAMES = {"vocab_file": "vocab.txt"}
|
14 |
-
|
15 |
-
# Define the mapping for pretrained vocabulary files
|
16 |
-
PRETRAINED_VOCAB_FILES_MAP = {
|
17 |
-
"vocab_file": {
|
18 |
-
"lca-mini-k6s1": "lca-base-dna6/vocab.txt",
|
19 |
-
"lca-mini-k6s2": "lca-base-dna6/vocab.txt",
|
20 |
-
"lca-mini-k1s1": "lca-base-dna1/vocab.txt",
|
21 |
-
}
|
22 |
-
}
|
23 |
-
|
24 |
-
# Define positional embedding sizes for pretrained models
|
25 |
-
PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES = {
|
26 |
-
"lca-mini-k6s1": 1024,
|
27 |
-
"lca-mini-k1s1": 1024,
|
28 |
-
"lca-mini-k6s2": 2048,
|
29 |
-
}
|
30 |
-
|
31 |
-
# Define initial configuration for pretrained models
|
32 |
-
PRETRAINED_INIT_CONFIGURATION = {
|
33 |
-
"lca-mini-k6s1": {"do_upper_case": True},
|
34 |
-
"lca-mini-k1s1": {"do_upper_case": True},
|
35 |
-
"lca-mini-k6s2": {"do_upper_case": True},
|
36 |
-
}
|
37 |
-
|
38 |
-
# Utility function to load vocabulary from a file
|
39 |
-
def load_vocab(vocab_file):
|
40 |
-
"""Loads a vocabulary file into a dictionary."""
|
41 |
-
vocab = collections.OrderedDict()
|
42 |
-
with open(vocab_file, "r", encoding="utf-8") as reader:
|
43 |
-
tokens = reader.readlines()
|
44 |
-
for index, token in enumerate(tokens):
|
45 |
-
vocab[token.rstrip("\n")] = index
|
46 |
-
return vocab
|
47 |
-
|
48 |
-
class LCATokenizer(PreTrainedTokenizer):
|
49 |
-
"""
|
50 |
-
Custom tokenizer for LCA (Local Context Aware) tasks.
|
51 |
-
Handles specific tokenization processes, including k-mer tokenization with configurable shifts.
|
52 |
-
|
53 |
-
Attributes:
|
54 |
-
vocab_files_names (dict): Mapping of vocabulary file names.
|
55 |
-
pretrained_vocab_files_map (dict): Mapping of pretrained vocabulary files.
|
56 |
-
pretrained_init_configuration (dict): Initial configuration for pretrained models.
|
57 |
-
max_model_input_sizes (dict): Maximum input sizes for pretrained models.
|
58 |
-
"""
|
59 |
-
|
60 |
-
vocab_files_names = VOCAB_FILES_NAMES
|
61 |
-
pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP
|
62 |
-
pretrained_init_configuration = PRETRAINED_INIT_CONFIGURATION
|
63 |
-
max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
|
64 |
-
|
65 |
-
nucleotide_abc = {"A", "T", "C", "G"}
|
66 |
-
extended_nucleotide_abc = {"A", "T", "C", "G", "*"}
|
67 |
-
sequence_unk_token = 'N'
|
68 |
-
|
69 |
-
default_unk_token = "[UNK]"
|
70 |
-
default_sep_token = "[SEP]"
|
71 |
-
default_pad_token = "[PAD]"
|
72 |
-
default_cls_token = "[CLS]"
|
73 |
-
default_mask_token = "[MASK]"
|
74 |
-
|
75 |
-
def __init__(
|
76 |
-
self,
|
77 |
-
config: Dict = {},
|
78 |
-
operation_space: str = "kmer",
|
79 |
-
**kwargs,
|
80 |
-
):
|
81 |
-
"""
|
82 |
-
Initializes the LCATokenizer with configuration and operation space.
|
83 |
-
|
84 |
-
Args:
|
85 |
-
config (dict): Tokenization parameters like k-mer size and shift.
|
86 |
-
operation_space (str): Defines operation mode ('kmer' or 'sequence').
|
87 |
-
kwargs: Additional arguments for PreTrainedTokenizer.
|
88 |
-
"""
|
89 |
-
self.defconfig = SeqConfig()
|
90 |
-
config = self.defconfig.get_and_set_tokenization_parameters(config)
|
91 |
-
self.config = config
|
92 |
-
self.operation_space = operation_space
|
93 |
-
|
94 |
-
# Set default tokens
|
95 |
-
kwargs.setdefault("cls_token", self.default_cls_token)
|
96 |
-
kwargs.setdefault("unk_token", self.default_unk_token)
|
97 |
-
kwargs.setdefault("sep_token", self.default_sep_token)
|
98 |
-
kwargs.setdefault("pad_token", self.default_pad_token)
|
99 |
-
kwargs.setdefault("mask_token", self.default_mask_token)
|
100 |
-
|
101 |
-
# Load vocabulary
|
102 |
-
vocab_file = self.config["vocabfile"]
|
103 |
-
self.vocab = self.config["vocabmap"]
|
104 |
-
self.id2token = {v: k for k, v in self.vocab.items()}
|
105 |
-
self.max_len = self.config["max_segment_length"]
|
106 |
-
|
107 |
-
super().__init__(**kwargs)
|
108 |
-
|
109 |
-
# Handle extended vocabulary for sequence mode
|
110 |
-
if self.operation_space == 'sequence':
|
111 |
-
token_extension = sorted(list(set(generate_kmers(LCATokenizer.extended_nucleotide_abc, self.config['kmer'])) - \
|
112 |
-
set(generate_kmers(LCATokenizer.nucleotide_abc, self.config['kmer'])) ))
|
113 |
-
self.extended_vocab = deepcopy(self.vocab)
|
114 |
-
for token in token_extension:
|
115 |
-
self.extended_vocab[token] = 4
|
116 |
-
|
117 |
-
self.unk_token = LCATokenizer.sequence_unk_token * self.config['shift']
|
118 |
-
self.mask_token = '*'
|
119 |
-
self.extended_vocab[self.mask_token] = self.vocab['[MASK]']
|
120 |
-
|
121 |
-
full_unk = 'N' * self.config['kmer']
|
122 |
-
self.vocab[full_unk] = 1
|
123 |
-
self.id2token[1] = full_unk
|
124 |
-
self.full_unk_token = full_unk
|
125 |
-
|
126 |
-
else:
|
127 |
-
self.extended_vocab = self.vocab
|
128 |
-
self.unk_token = '[UNK]'
|
129 |
-
|
130 |
-
self.unkown_tokenid = self.vocab['[UNK]']
|
131 |
-
self.sep_token = '[SEP]'
|
132 |
-
self.cls_token = '[CLS]'
|
133 |
-
self.pad_token = '[PAD]'
|
134 |
-
self.mask_token = '[MASK]'
|
135 |
-
self.special_tokens = list(self.special_tokens_map.values())
|
136 |
-
|
137 |
-
|
138 |
-
|
139 |
-
def _tokenize(self, text, **kwargs):
|
140 |
-
"""
|
141 |
-
Tokenizes the input text using LCA tokenization with an optional offset.
|
142 |
-
|
143 |
-
Args:
|
144 |
-
text (str): The input DNA sequence to tokenize.
|
145 |
-
kwargs: Additional arguments, including:
|
146 |
-
- offset (int): The starting position for tokenization. Default is 0.
|
147 |
-
|
148 |
-
Returns:
|
149 |
-
List[str]: A list of tokens generated from the input text.
|
150 |
-
"""
|
151 |
-
offset = kwargs.get("offset", 0)
|
152 |
-
#if offset < 0 or offset >= self.config.get("shift", 1):
|
153 |
-
# raise ValueError(f"Invalid offset: {offset}. Must be between 0 and {self.config['shift'] - 1}.")
|
154 |
-
|
155 |
-
return lca_kmer_tokenize_segment(text, offset, self.config)
|
156 |
-
|
157 |
-
def _convert_token_to_id(self, token: str) -> int:
|
158 |
-
"""
|
159 |
-
Converts a token to its corresponding ID using the vocabulary.
|
160 |
-
|
161 |
-
Args:
|
162 |
-
token (str): The token to convert.
|
163 |
-
|
164 |
-
Returns:
|
165 |
-
int: Token ID, or the unknown token ID if the token is not in the vocabulary.
|
166 |
-
"""
|
167 |
-
return self.extended_vocab.get(token, self.unkown_tokenid)
|
168 |
-
|
169 |
-
def _convert_id_to_token(self, index: int) -> str:
|
170 |
-
"""
|
171 |
-
Converts an ID to its corresponding token using the vocabulary.
|
172 |
-
|
173 |
-
Args:
|
174 |
-
index (int): The ID to convert.
|
175 |
-
|
176 |
-
Returns:
|
177 |
-
str: Corresponding token, or the unknown token if the ID is not in the vocabulary.
|
178 |
-
"""
|
179 |
-
|
180 |
-
|
181 |
-
return self.id2token.get(index, self.unk_token)
|
182 |
-
|
183 |
-
def __len__(self) -> int:
|
184 |
-
"""
|
185 |
-
Returns the length of the tokenizer's vocabulary.
|
186 |
-
|
187 |
-
The length returned is one less than the actual number of items in the vocabulary
|
188 |
-
to account for a specific offset or adjustment in token indexing.
|
189 |
-
|
190 |
-
:return: The adjusted length of the vocabulary.
|
191 |
-
:rtype: int
|
192 |
-
"""
|
193 |
-
return len(self.vocab)
|
194 |
-
|
195 |
-
|
196 |
-
|
197 |
-
def tokenize(self, text: str, **kwargs) -> List[str]:
|
198 |
-
"""
|
199 |
-
Tokenizes the input text using LCA tokenization.
|
200 |
-
|
201 |
-
Args:
|
202 |
-
text (str): The input DNA sequence to tokenize.
|
203 |
-
kwargs: Additional arguments, including:
|
204 |
-
- offset (int): The starting position for tokenization. Default is 0.
|
205 |
-
|
206 |
-
Returns:
|
207 |
-
List[str]: A list of tokens generated from the input text.
|
208 |
-
"""
|
209 |
-
return self._tokenize(text, **kwargs)
|
210 |
-
|
211 |
-
def encode(self, text: str, **kwargs) -> List[int]:
|
212 |
-
"""
|
213 |
-
Extends the base `encode` method to support an `offset` parameter for custom tokenization logic.
|
214 |
-
|
215 |
-
Args:
|
216 |
-
text (str): Input text (DNA sequence).
|
217 |
-
offset (int): Offset parameter for the LCA tokenization. Defaults to 0.
|
218 |
-
kwargs: Additional arguments passed to the base `encode` method.
|
219 |
-
|
220 |
-
Returns:
|
221 |
-
List[int]: Encoded token IDs.
|
222 |
-
"""
|
223 |
-
# Inject the offset into kwargs for the tokenizer
|
224 |
-
offset = kwargs.get("offset", 0)
|
225 |
-
kwargs["offset"] = offset
|
226 |
-
return super().encode(text, **kwargs)
|
227 |
-
|
228 |
-
def build_inputs_with_special_tokens(
|
229 |
-
self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
|
230 |
-
) -> List[int]:
|
231 |
-
"""
|
232 |
-
Builds inputs by adding special tokens to a sequence or pair of sequences.
|
233 |
-
|
234 |
-
Args:
|
235 |
-
token_ids_0 (List[int]): List of token IDs for the first sequence.
|
236 |
-
token_ids_1 (List[int], optional): List of token IDs for the second sequence.
|
237 |
-
|
238 |
-
Returns:
|
239 |
-
List[int]: Input IDs with special tokens.
|
240 |
-
"""
|
241 |
-
if token_ids_1 is None:
|
242 |
-
return [self.cls_token_id] + token_ids_0 + [self.sep_token_id]
|
243 |
-
|
244 |
-
input_ids = [self.cls_token_id] + token_ids_0 + [self.sep_token_id] + token_ids_1 + [self.sep_token_id]
|
245 |
-
#token_type_ids = [0 for i in range(len(input_ids))]
|
246 |
-
return input_ids
|
247 |
-
|
248 |
-
def create_token_type_ids_from_sequences(
|
249 |
-
self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
|
250 |
-
) -> List[int]:
|
251 |
-
"""
|
252 |
-
Create the token type IDs corresponding to the sequences passed. [What are token type
|
253 |
-
IDs?](../glossary#token-type-ids)
|
254 |
-
|
255 |
-
Should be overridden in a subclass if the model has a special way of building those.
|
256 |
-
|
257 |
-
Args:
|
258 |
-
token_ids_0 (`List[int]`): The first tokenized sequence.
|
259 |
-
token_ids_1 (`List[int]`, *optional*): The second tokenized sequence.
|
260 |
-
|
261 |
-
Returns:
|
262 |
-
`List[int]`: The token type ids.
|
263 |
-
"""
|
264 |
-
if token_ids_1 is None:
|
265 |
-
return (len(token_ids_0)+2) * [0]
|
266 |
-
return [0] * len(token_ids_0) + [1] * len(token_ids_1)
|
267 |
-
|
268 |
-
def batch_encode_plus(self, *args, **kwargs):
|
269 |
-
"""
|
270 |
-
Extends the base `batch_encode_plus` method to add custom functionality if needed.
|
271 |
-
|
272 |
-
Args:
|
273 |
-
*args: Positional arguments passed to the base method.
|
274 |
-
**kwargs: Keyword arguments passed to the base method.
|
275 |
-
|
276 |
-
Returns:
|
277 |
-
dict: A dictionary containing the results of batch encoding.
|
278 |
-
"""
|
279 |
-
# Call the parent method to handle the batch encoding
|
280 |
-
#print('Running batch encoding with ids')
|
281 |
-
act_outputs = super().batch_encode_plus(*args, **kwargs)
|
282 |
-
return act_outputs
|
283 |
-
|
284 |
-
|
285 |
-
def save_vocabulary(self, save_directory: str, filename_prefix: Optional[str] = None) -> Tuple[str]:
|
286 |
-
"""
|
287 |
-
Saves the tokenizer's vocabulary to a file.
|
288 |
-
|
289 |
-
Args:
|
290 |
-
save_directory (str): Directory to save the vocabulary file.
|
291 |
-
filename_prefix (str, optional): Prefix for the filename. Default is None.
|
292 |
-
|
293 |
-
Returns:
|
294 |
-
Tuple[str]: Path to the saved vocabulary file.
|
295 |
-
"""
|
296 |
-
if filename_prefix is None:
|
297 |
-
filename_prefix = ""
|
298 |
-
vocab_file_path = os.path.join(save_directory, filename_prefix + "vocab.txt")
|
299 |
-
with open(vocab_file_path, "w") as f:
|
300 |
-
for token in self.vocab:
|
301 |
-
f.write(token + "\n")
|
302 |
-
return (vocab_file_path,)
|
303 |
-
|
304 |
-
def save_pretrained(self, save_directory: str, **kwargs):
|
305 |
-
"""
|
306 |
-
Saves the tokenizer configuration and vocabulary to a directory.
|
307 |
-
|
308 |
-
Args:
|
309 |
-
save_directory (str): Directory to save the tokenizer files.
|
310 |
-
"""
|
311 |
-
if not os.path.exists(save_directory):
|
312 |
-
os.makedirs(save_directory)
|
313 |
-
super().save_pretrained(save_directory, **kwargs)
|
314 |
-
|
315 |
-
tokenizer_config_path = os.path.join(save_directory, "tokenizer_config.json")
|
316 |
-
if os.path.exists(tokenizer_config_path):
|
317 |
-
with open(tokenizer_config_path, "r") as f:
|
318 |
-
tokenizer_config = json.load(f)
|
319 |
-
else:
|
320 |
-
tokenizer_config = {}
|
321 |
-
|
322 |
-
tokenizer_config.update({
|
323 |
-
"kmer": self.config.get("kmer", 6),
|
324 |
-
"shift": self.config.get("shift", 1),
|
325 |
-
})
|
326 |
-
|
327 |
-
with open(tokenizer_config_path, "w") as f:
|
328 |
-
json.dump(tokenizer_config, f, indent=2)
|
329 |
-
|
330 |
-
@classmethod
|
331 |
-
def from_pretrained(cls, pretrained_model_name_or_path, **kwargs):
|
332 |
-
"""
|
333 |
-
Loads a tokenizer from the pretrained model directory or Hugging Face Hub.
|
334 |
-
|
335 |
-
Args:
|
336 |
-
pretrained_model_name_or_path (str): Path or model name on Hugging Face Hub.
|
337 |
-
kwargs: Additional arguments for initialization.
|
338 |
-
|
339 |
-
Returns:
|
340 |
-
LCATokenizer: The loaded tokenizer instance.
|
341 |
-
"""
|
342 |
-
tokenizer_config_file = hf_hub_url(
|
343 |
-
pretrained_model_name_or_path, filename="tokenizer_config.json"
|
344 |
-
)
|
345 |
-
resolved_tokenizer_config_file = cached_file(
|
346 |
-
pretrained_model_name_or_path, filename="tokenizer_config.json"
|
347 |
-
)
|
348 |
-
|
349 |
-
with open(resolved_tokenizer_config_file, "r") as f:
|
350 |
-
tokenizer_config = json.load(f)
|
351 |
-
|
352 |
-
kmer = tokenizer_config.pop("kmer", 6)
|
353 |
-
shift = tokenizer_config.pop("shift", 1)
|
354 |
-
base_tokenization_config = {'kmer': kmer, 'shift': shift}
|
355 |
-
defconfig = SeqConfig()
|
356 |
-
config = defconfig.get_and_set_tokenization_parameters(base_tokenization_config)
|
357 |
-
|
358 |
-
tokenizer = super().from_pretrained(pretrained_model_name_or_path, **kwargs)
|
359 |
-
tokenizer.config = config
|
360 |
-
|
361 |
-
return tokenizer
|
362 |
-
|
363 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
tokenizer_config.json
CHANGED
@@ -1,10 +1,4 @@
|
|
1 |
{
|
2 |
-
"auto_map": {
|
3 |
-
"AutoTokenizer": [
|
4 |
-
"tokenizer.LCATokenizer",
|
5 |
-
null
|
6 |
-
]
|
7 |
-
},
|
8 |
"clean_up_tokenization_spaces": true,
|
9 |
"cls_token": "[CLS]",
|
10 |
"mask_token": "[MASK]",
|
@@ -15,4 +9,4 @@
|
|
15 |
"unk_token": "[UNK]",
|
16 |
"kmer": 6,
|
17 |
"shift": 1
|
18 |
-
}
|
|
|
1 |
{
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
"clean_up_tokenization_spaces": true,
|
3 |
"cls_token": "[CLS]",
|
4 |
"mask_token": "[MASK]",
|
|
|
9 |
"unk_token": "[UNK]",
|
10 |
"kmer": 6,
|
11 |
"shift": 1
|
12 |
+
}
|