forked from cgpotts/cs224u
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtest_vsm.py
265 lines (221 loc) · 7.14 KB
/
test_vsm.py
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
import numpy as np
import os
import pandas as pd
import pytest
import torch
from transformers import BertModel, BertTokenizer
import vsm
import utils
__author__ = "Christopher Potts"
__version__ = "CS224u, Stanford, Spring 2022"
utils.fix_random_seeds()
DATA_HOME = os.path.join('data', 'vsmdata')
REL_HOME = os.path.join('data', 'wordrelatedness')
@pytest.fixture
def df():
vocab = ['ab', 'bc', 'cd', 'de']
df = pd.DataFrame([
[ 4., 4., 2., 0.],
[ 4., 61., 8., 18.],
[ 2., 8., 10., 0.],
[ 0., 18., 0., 5.]],
index=vocab, columns=vocab)
return df
@pytest.fixture
def count_dfs():
dfs = {}
basenames = (
'yelp_window5-scaled.csv.gz',
'yelp_window20-flat.csv.gz',
'giga_window5-scaled.csv.gz',
'giga_window20-flat.csv.gz')
for basename in basenames:
dfs[basename] = pd.read_csv(
os.path.join(DATA_HOME, basename), index_col=0)
return dfs
@pytest.mark.parametrize("arg, expected", [
[
np.array([[34.0, 11.0], [ 47.0, 7.0]]),
np.array([[0.92345679, 1.34444444], [1.06378601, 0.71296296]])
]
])
def test_observed_over_expected(arg, expected):
result = vsm.observed_over_expected(arg).round(8)
assert np.array_equal(result, expected.round(8))
@pytest.mark.parametrize("arg, expected, positive", [
[
np.array([[34.0, 11.0], [ 47.0, 7.0]]),
np.array([[0.0, 0.29598088], [0.06183425, 0.0]]),
True
],
[
np.array([[34.0, 11.0], [ 47.0, 7.0]]),
np.array([[-0.07963127, 0.29598088], [0.06183425, -0.33832581]]),
False
],
[
np.array([[1.0, 0.0, 0.0], [1000.0, 1000.0, 4000.0], [1000.0, 2000.0, 999.0]]),
np.array([[1.60893804, 0., 0.],[0., 0., 0.28788209], [0.22289371, 0.51107566, 0.]]),
True
]
])
def test_pmi(arg, expected, positive):
result = vsm.pmi(arg, positive=positive).round(8)
assert np.array_equal(result, expected.round(8))
@pytest.mark.parametrize("arg, expected", [
[
np.array([[34.0, 0.0], [0.0, 7.0]]),
np.array([[0.69314718, 0.0], [0.0, 0.69314718]])
],
[
np.array([
[10.0, 10.0, 10.0, 10.0],
[10.0, 10.0, 10.0, 0.0],
[10.0, 10.0, 0.0, 0.0],
[0.0, 0.0, 0.0, 1.0]]),
np.array([
[0.0, 0.0, 0.0, 0.0],
[0.09589402, 0.09589402, 0.14384104, 0.],
[0.23104906, 0.23104906, 0., 0.],
[0., 0., 0., 0.12602676]])
]
])
def test_tfidf(arg, expected):
result = vsm.tfidf(arg).round(8)
expected = expected.round(8)
assert np.array_equal(result, expected)
@pytest.mark.parametrize("bigram, expected", [
[
'<w>a', np.array([4.0, 4.0, 2.0, 0.0])
],
[
'<w>b', np.array([4.0, 61.0, 8.0, 18.0])
],
])
def test_ngram_vsm(df, bigram, expected):
X = vsm.ngram_vsm(df)
result = X.loc[bigram]
assert np.array_equal(result, expected)
@pytest.mark.parametrize("word, n, expected", [
['abc', 1, ['a', 'b', 'c']],
['abc', 2, ['<w>a', 'ab', 'bc', 'c</w>']]
])
def test_get_character_ngrams(word, n, expected):
result = vsm.get_character_ngrams(word, n=n)
assert result == expected
def test_tsne_viz(df):
vsm.tsne_viz(df)
def test_lsa(df):
vsm.lsa(df, k=2)
@pytest.mark.parametrize("basename1, basename2", [
['yelp_window5-scaled.csv.gz', 'yelp_window20-flat.csv.gz'],
['yelp_window20-flat.csv.gz', 'giga_window5-scaled.csv.gz'],
['giga_window5-scaled.csv.gz', 'giga_window20-flat.csv.gz']
])
def test_count_matrix_index_identity(basename1, basename2, count_dfs):
df1 = count_dfs[basename1]
df2 = count_dfs[basename2]
assert df1.index.equals(df2.index)
@pytest.mark.parametrize("batch_ids, expected_shape", [
(
[[101, 14324, 102]], (1, 3, 768)
)
])
def test_hf_represent(batch_ids, expected_shape):
"""Really just tests that the function works."""
batch_ids = torch.LongTensor(batch_ids)
model = BertModel.from_pretrained('bert-base-uncased')
reps = vsm.hf_represent(batch_ids, model, layer=-1)
assert reps.shape == expected_shape
@pytest.mark.parametrize("text, add_special_tokens, expected_len", [
("the cat", True, 4),
("the cat", False, 2)
])
def test_hf_encode(text, add_special_tokens, expected_len):
"""Really just tests that the function works."""
tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
encoding = vsm.hf_encode(
text, tokenizer, add_special_tokens=add_special_tokens)
assert encoding.shape == (1, expected_len)
@pytest.mark.parametrize("X, expected", [
([[[1., 2, 3], [4., 5, 6]]], [[2.5, 3.5, 4.5]])
])
def test_mean_pooling(X, expected):
X = torch.tensor(X)
expected = torch.tensor(expected)
result = vsm.mean_pooling(X)
assert torch.equal(result, expected)
@pytest.mark.parametrize("X, expected", [
[
[[[1., 4, 3], [4., 2, 6]]],
[[4., 4, 6]]
],
[
[[[1., 4, 3], [4., 2, 6]], [[1., 4, 3], [4., 2, 6]]],
[[4., 4, 6], [4., 4, 6]]
],
])
def test_max_pooling(X, expected):
X = torch.tensor(X)
expected = torch.tensor(expected)
result = vsm.max_pooling(X)
assert torch.equal(result, expected)
@pytest.mark.parametrize("X, expected", [
[
[[[1., 4, 3], [4., 2, 6]]],
[[1., 2, 3]]
],
[
[[[1., 4, 3], [4., 2, 6]], [[4, 3, 2], [2, 3, 4.]]],
[[1., 2, 3], [2., 3, 2]]
]
])
def test_min_pooling(X, expected):
X = torch.tensor(X)
expected = torch.tensor(expected)
result = vsm.min_pooling(X)
assert torch.equal(result, expected)
@pytest.mark.parametrize("X, expected", [
[
[[[1., 4, 3], [4., 2, 6]]],
[[4., 2, 6]]
],
[
[[[1., 4, 3], [4., 2, 6]], [[4, 3, 2], [2, 3, 4.]]],
[[4., 2, 6], [2., 3, 4]]
]
])
def test_last_pooling(X, expected):
X = torch.tensor(X)
expected = torch.tensor(expected)
result = vsm.last_pooling(X)
assert torch.equal(result, expected)
@pytest.mark.parametrize("pool_func", [
vsm.mean_pooling,
vsm.max_pooling,
vsm.min_pooling,
vsm.last_pooling
])
def test_pool_func_shape_check_raise(pool_func):
hidden_states = torch.Tensor([1,2])
with pytest.raises(ValueError):
pool_func(hidden_states)
def test_create_subword_pooling_vsm():
"""Really just tests that the function works."""
vocab = ["puppy", "snuffleupagus"]
bert_weights_name = 'bert-base-uncased'
tokenizer = BertTokenizer.from_pretrained(bert_weights_name)
model = BertModel.from_pretrained(bert_weights_name)
df = vsm.create_subword_pooling_vsm(
vocab, tokenizer, model,
layer=1, pool_func=vsm.mean_pooling)
assert list(df.index) == vocab
def test_word_relatedness_evaluation():
"""Really just tests that the function works."""
dev_df = pd.read_csv(
os.path.join(REL_HOME, "cs224u-wordrelatedness-dev.csv"))
count_df = pd.read_csv(
os.path.join(DATA_HOME, "giga_window5-scaled.csv.gz"), index_col=0)
count_pred_df, count_rho = vsm.word_relatedness_evaluation(dev_df, count_df)
assert isinstance(count_rho, float)
assert 'prediction' in count_pred_df.columns