File size: 8,741 Bytes
158b61b |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 |
"""Base Transform class and relate utils."""
import torch
from onmt.utils.logging import logger
from onmt.utils.misc import check_path
from onmt.inputters.fields import get_vocabs
class Transform(object):
"""A Base class that every transform method should derived from."""
def __init__(self, opts):
"""Initialize Transform by parsing `opts` and add them as attribute."""
self.opts = opts
self._parse_opts()
def _set_seed(self, seed):
"""Reproducibility: Set seed for non-deterministic transform."""
pass
@classmethod
def require_vocab(cls):
"""Override this method to inform it need vocab to start."""
return False
def warm_up(self, vocabs=None):
"""Procedure needed after initialize and before apply.
This should be override if there exist any procedure necessary
before `apply`, like setups based on parsed options or load models,
etc.
"""
if self.opts.seed > 0:
self._set_seed(self.opts.seed)
if self.require_vocab():
if vocabs is None:
raise ValueError(f"{type(self).__name__} requires vocabs!")
self.vocabs = vocabs
@classmethod
def add_options(cls, parser):
"""Available options relate to this Transform."""
pass
@classmethod
def _validate_options(cls, opts):
"""Extra checks to validate options added from `add_options`."""
pass
@classmethod
def get_specials(cls, opts):
return (set(), set())
def apply(self, example, is_train=False, stats=None, **kwargs):
"""Apply transform to `example`.
Args:
example (dict): a dict of field value, ex. src, tgt;
is_train (bool): Indicate if src/tgt is training data;
stats (TransformStatistics): a statistic object.
"""
raise NotImplementedError
def apply_reverse(self, translated):
return translated
def __getstate__(self):
"""Pickling following for rebuild."""
state = {"opts": self.opts}
if hasattr(self, 'vocabs'):
state['vocabs'] = self.vocabs
return state
def _parse_opts(self):
"""Parse opts to set/reset instance's attributes.
This should be override if there are attributes other than self.opts.
To make sure we recover from picked state.
(This should only contain attribute assignment, other routine is
suggest to define in `warm_up`.)
"""
pass
def __setstate__(self, state):
"""Reload when unpickling from save file."""
self.opts = state["opts"]
self._parse_opts()
vocabs = state.get('vocabs', None)
self.warm_up(vocabs=vocabs)
def stats(self):
"""Return statistic message."""
return ''
def _repr_args(self):
"""Return str represent key arguments for class."""
return ''
def __repr__(self):
cls_name = type(self).__name__
cls_args = self._repr_args()
return '{}({})'.format(cls_name, cls_args)
class ObservableStats:
"""A running observable statistics."""
__slots__ = []
def name(self) -> str:
"""Return class name as name for statistics."""
return type(self).__name__
def update(self, other: "ObservableStats"):
"""Update current statistics with others."""
raise NotImplementedError
def __str__(self) -> str:
return "{}({})".format(
self.name(),
", ".join(
f"{name}={getattr(self, name)}" for name in self.__slots__
)
)
class TransformStatistics:
"""A observer containing runing statistics."""
def __init__(self):
self.observables = {}
def update(self, observable: ObservableStats):
"""Adding observable to observe/updating existing observable."""
name = observable.name()
if name not in self.observables:
self.observables[name] = observable
else:
self.observables[name].update(observable)
def report(self):
"""Pop out all observing statistics and reporting them."""
msgs = []
report_ids = list(self.observables.keys())
for name in report_ids:
observable = self.observables.pop(name)
msgs.append(f"\t\t\t* {str(observable)}")
if len(msgs) != 0:
return "\n".join(msgs)
else:
return ""
class TransformPipe(Transform):
"""Pipeline built by a list of Transform instance."""
def __init__(self, opts, transform_list):
"""Initialize pipeline by a list of transform instance."""
self.opts = None # opts is not required
self.transforms = transform_list
self.statistics = TransformStatistics()
@classmethod
def build_from(cls, transform_list):
"""Return a `TransformPipe` instance build from `transform_list`."""
for transform in transform_list:
assert isinstance(transform, Transform), \
"transform should be a instance of Transform."
transform_pipe = cls(None, transform_list)
return transform_pipe
def warm_up(self, vocabs):
"""Warm up Pipeline by iterate over all transfroms."""
for transform in self.transforms:
transform.warm_up(vocabs)
@classmethod
def get_specials(cls, opts, transforms):
"""Return all specials introduced by `transforms`."""
src_specials, tgt_specials = set(), set()
for transform in transforms:
_src_special, _tgt_special = transform.get_specials(transform.opts)
src_specials.update(_src_special)
tgt_specials.update(tgt_specials)
return (src_specials, tgt_specials)
def apply(self, example, is_train=False, **kwargs):
"""Apply transform pipe to `example`.
Args:
example (dict): a dict of field value, ex. src, tgt.
"""
for transform in self.transforms:
example = transform.apply(
example, is_train=is_train, stats=self.statistics, **kwargs)
if example is None:
break
return example
def apply_reverse(self, translated):
for transform in self.transforms:
translated = transform.apply_reverse(translated)
return translated
def __getstate__(self):
"""Pickling following for rebuild."""
return (self.opts, self.transforms, self.statistics)
def __setstate__(self, state):
"""Reload when unpickling from save file."""
self.opts, self.transforms, self.statistics = state
def stats(self):
"""Return statistic message."""
return self.statistics.report()
def _repr_args(self):
"""Return str represent key arguments for class."""
info_args = []
for transform in self.transforms:
info_args.append(repr(transform))
return ', '.join(info_args)
def make_transforms(opts, transforms_cls, fields):
"""Build transforms in `transforms_cls` with vocab of `fields`."""
vocabs = get_vocabs(fields) if fields is not None else None
transforms = {}
for name, transform_cls in transforms_cls.items():
if transform_cls.require_vocab() and vocabs is None:
logger.warning(
f"{transform_cls.__name__} require vocab to apply, skip it."
)
continue
transform_obj = transform_cls(opts)
transform_obj.warm_up(vocabs)
transforms[name] = transform_obj
return transforms
def get_specials(opts, transforms_cls_dict):
"""Get specials of transforms that should be registed in Vocab."""
all_specials = {'src': set(), 'tgt': set()}
for name, transform_cls in transforms_cls_dict.items():
src_specials, tgt_specials = transform_cls.get_specials(opts)
all_specials['src'].update(src_specials)
all_specials['tgt'].update(tgt_specials)
logger.info(f"Get special vocabs from Transforms: {all_specials}.")
return all_specials
def save_transforms(transforms, save_data, overwrite=True):
"""Dump `transforms` object."""
transforms_path = "{}.transforms.pt".format(save_data)
check_path(transforms_path, exist_ok=overwrite, log=logger.warning)
logger.info(f"Saving Transforms to {transforms_path}.")
torch.save(transforms, transforms_path)
def load_transforms(opts):
"""Load dumped `transforms` object."""
transforms_path = "{}.transforms.pt".format(opts.save_data)
transforms = torch.load(transforms_path)
logger.info("Transforms loaded.")
return transforms
|