Jon Gauthier
commited on
Commit
·
7e951ff
1
Parent(s):
2b2f744
working example with half-implemented evaluation
Browse files- syntaxgym/prediction.py +8 -1
- syntaxgym/syntaxgym.py +51 -6
- test.py +12 -0
syntaxgym/prediction.py
CHANGED
@@ -4,7 +4,14 @@ from typing import Union, Optional as TOptional, List as TList
|
|
4 |
from pyparsing import *
|
5 |
import numpy as np
|
6 |
|
7 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
8 |
|
9 |
|
10 |
# Enable parser packrat (caching)
|
|
|
4 |
from pyparsing import *
|
5 |
import numpy as np
|
6 |
|
7 |
+
METRICS = {
|
8 |
+
'sum': sum,
|
9 |
+
'mean': np.mean,
|
10 |
+
'median': np.median,
|
11 |
+
'range': np.ptp,
|
12 |
+
'max': max,
|
13 |
+
'min': min
|
14 |
+
}
|
15 |
|
16 |
|
17 |
# Enable parser packrat (caching)
|
syntaxgym/syntaxgym.py
CHANGED
@@ -5,12 +5,16 @@ SyntaxGym dataset as used in Hu et al. (2020).
|
|
5 |
"""
|
6 |
|
7 |
|
|
|
8 |
import json
|
9 |
from pathlib import Path
|
|
|
10 |
from typing import List
|
11 |
|
12 |
import datasets
|
13 |
|
|
|
|
|
14 |
|
15 |
_CITATION = """
|
16 |
@inproceedings{Hu:et-al:2020,
|
@@ -28,11 +32,11 @@ _PROJECT_URL = "https://syntaxgym.org"
|
|
28 |
_DOWNLOAD_URL = "https://github.com/cpllab/syntactic-generalization"
|
29 |
|
30 |
|
31 |
-
print("_____HERE")
|
32 |
SUITE_JSONS = []
|
33 |
for suite_f in Path("test_suites").glob("*.json"):
|
34 |
with suite_f.open() as f:
|
35 |
SUITE_JSONS.append(json.load(f))
|
|
|
36 |
|
37 |
|
38 |
class SyntaxGymSuiteConfig(datasets.BuilderConfig):
|
@@ -51,11 +55,12 @@ class SyntaxGymSuiteConfig(datasets.BuilderConfig):
|
|
51 |
class SyntaxGym(datasets.GeneratorBasedBuilder):
|
52 |
|
53 |
BUILDER_CONFIGS = [SyntaxGymSuiteConfig(suite_json)
|
54 |
-
for suite_json in SUITE_JSONS]
|
55 |
|
56 |
def _info(self):
|
57 |
condition_spec = {
|
58 |
"condition_name": datasets.Value("string"),
|
|
|
59 |
"regions": datasets.Sequence({
|
60 |
"region_number": datasets.Value("int32"),
|
61 |
"content": datasets.Value("string")
|
@@ -85,10 +90,50 @@ class SyntaxGym(datasets.GeneratorBasedBuilder):
|
|
85 |
|
86 |
def _generate_examples(self, name):
|
87 |
# DEV: NB suite jsons already loaded because BUILDER_CONFIGS is static
|
88 |
-
|
89 |
-
|
90 |
-
suite_json = next(suite for suite in SUITE_JSONS
|
91 |
-
if suite["meta"]["name"] == name)
|
92 |
|
93 |
for item in suite_json["items"]:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
94 |
yield item["item_number"], item
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
5 |
"""
|
6 |
|
7 |
|
8 |
+
from copy import deepcopy
|
9 |
import json
|
10 |
from pathlib import Path
|
11 |
+
import re
|
12 |
from typing import List
|
13 |
|
14 |
import datasets
|
15 |
|
16 |
+
from syntaxgym.prediction import Prediction
|
17 |
+
|
18 |
|
19 |
_CITATION = """
|
20 |
@inproceedings{Hu:et-al:2020,
|
|
|
32 |
_DOWNLOAD_URL = "https://github.com/cpllab/syntactic-generalization"
|
33 |
|
34 |
|
|
|
35 |
SUITE_JSONS = []
|
36 |
for suite_f in Path("test_suites").glob("*.json"):
|
37 |
with suite_f.open() as f:
|
38 |
SUITE_JSONS.append(json.load(f))
|
39 |
+
SUITE_JSONS = {suite["meta"]["name"]: suite for suite in SUITE_JSONS}
|
40 |
|
41 |
|
42 |
class SyntaxGymSuiteConfig(datasets.BuilderConfig):
|
|
|
55 |
class SyntaxGym(datasets.GeneratorBasedBuilder):
|
56 |
|
57 |
BUILDER_CONFIGS = [SyntaxGymSuiteConfig(suite_json)
|
58 |
+
for suite_json in SUITE_JSONS.values()]
|
59 |
|
60 |
def _info(self):
|
61 |
condition_spec = {
|
62 |
"condition_name": datasets.Value("string"),
|
63 |
+
"content": datasets.Value("string"),
|
64 |
"regions": datasets.Sequence({
|
65 |
"region_number": datasets.Value("int32"),
|
66 |
"content": datasets.Value("string")
|
|
|
90 |
|
91 |
def _generate_examples(self, name):
|
92 |
# DEV: NB suite jsons already loaded because BUILDER_CONFIGS is static
|
93 |
+
suite_json = SUITE_JSONS[name]
|
|
|
|
|
|
|
94 |
|
95 |
for item in suite_json["items"]:
|
96 |
+
# Convert to sentence input.
|
97 |
+
for cond in item["conditions"]:
|
98 |
+
cond["content"] = " ".join([region["content"].lstrip()
|
99 |
+
for region in cond["regions"]
|
100 |
+
if region["content"].strip() != ""])
|
101 |
+
cond["content"] = re.sub(r"\s+,", ",", cond["content"])
|
102 |
+
|
103 |
yield item["item_number"], item
|
104 |
+
|
105 |
+
|
106 |
+
class SyntaxGymMetric(datasets.Metric):
|
107 |
+
"""
|
108 |
+
SyntaxGym prediction evaluation metric.
|
109 |
+
"""
|
110 |
+
|
111 |
+
def __init__(self, *args, **kwargs):
|
112 |
+
super().__init__(*args, **kwargs)
|
113 |
+
self.suite = SUITE_JSONS[self.config_name]
|
114 |
+
self.predictions = [
|
115 |
+
Prediction(idx, p["formula"], "sum")
|
116 |
+
for idx, p in enumerate(self.suite["predictions"])
|
117 |
+
]
|
118 |
+
|
119 |
+
def _info(self):
|
120 |
+
features = datasets.Features({
|
121 |
+
"conditions": datasets.Sequence({
|
122 |
+
"condition_name": datasets.Value("string"),
|
123 |
+
"regions": datasets.Sequence({
|
124 |
+
"region_number": datasets.Value("int32"),
|
125 |
+
"metric_value": {
|
126 |
+
"sum": datasets.Value("float32")
|
127 |
+
},
|
128 |
+
}),
|
129 |
+
})
|
130 |
+
})
|
131 |
+
return datasets.MetricInfo(
|
132 |
+
description="TODO",
|
133 |
+
citation=_CITATION,
|
134 |
+
inputs_description="TODO",
|
135 |
+
features=features,
|
136 |
+
)
|
137 |
+
|
138 |
+
def _compute(self, region_surprisals):
|
139 |
+
suite_with_results = deepcopy(self.suite)
|
test.py
CHANGED
@@ -1,4 +1,16 @@
|
|
1 |
import datasets
|
|
|
2 |
|
3 |
|
4 |
dataset = datasets.load_dataset("syntaxgym", "mvrr_mod")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import datasets
|
2 |
+
import transformers
|
3 |
|
4 |
|
5 |
dataset = datasets.load_dataset("syntaxgym", "mvrr_mod")
|
6 |
+
metric = datasets.load_metric("syntaxgym", "mvrr_mod")
|
7 |
+
|
8 |
+
model = transformers.AutoModelForCausalLM.from_pretrained("gpt2")
|
9 |
+
|
10 |
+
|
11 |
+
for item in dataset["test"]:
|
12 |
+
# TODO evaluate surprisals
|
13 |
+
|
14 |
+
print(item)
|
15 |
+
|
16 |
+
break
|