-
Notifications
You must be signed in to change notification settings - Fork 58
/
test_funksvd.py
220 lines (163 loc) · 6.57 KB
/
test_funksvd.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
import logging
import pickle
from pathlib import Path
import lenskit.algorithms.funksvd as svd
import pandas as pd
import numpy as np
from pytest import approx, mark
import lenskit.util.test as lktu
_log = logging.getLogger(__name__)
simple_df = pd.DataFrame({'item': [1, 1, 2, 3],
'user': [10, 12, 10, 13],
'rating': [4.0, 3.0, 5.0, 2.0]})
def test_fsvd_basic_build():
algo = svd.FunkSVD(20, iterations=20)
algo.fit(simple_df)
assert algo.global_bias_ == approx(simple_df.rating.mean())
assert algo.item_features_.shape == (3, 20)
assert algo.user_features_.shape == (3, 20)
def test_fsvd_clamp_build():
algo = svd.FunkSVD(20, iterations=20, range=(1, 5))
algo.fit(simple_df)
assert algo.global_bias_ == approx(simple_df.rating.mean())
assert algo.item_features_.shape == (3, 20)
assert algo.user_features_.shape == (3, 20)
def test_fsvd_predict_basic():
algo = svd.FunkSVD(20, iterations=20)
algo.fit(simple_df)
assert algo.global_bias_ == approx(simple_df.rating.mean())
assert algo.item_features_.shape == (3, 20)
assert algo.user_features_.shape == (3, 20)
preds = algo.predict_for_user(10, [3])
assert len(preds) == 1
assert preds.index[0] == 3
assert preds.loc[3] >= 0
assert preds.loc[3] <= 5
def test_fsvd_predict_clamp():
algo = svd.FunkSVD(20, iterations=20, range=(1, 5))
algo.fit(simple_df)
assert algo.global_bias_ == approx(simple_df.rating.mean())
assert algo.item_features_.shape == (3, 20)
assert algo.user_features_.shape == (3, 20)
preds = algo.predict_for_user(10, [3])
assert len(preds) == 1
assert preds.index[0] == 3
assert preds.loc[3] >= 1
assert preds.loc[3] <= 5
def test_fsvd_no_bias():
algo = svd.FunkSVD(20, iterations=20, bias=None)
algo.fit(simple_df)
assert algo.global_bias_ == 0
assert algo.item_bias_ is None
assert algo.user_bias_ is None
assert algo.item_features_.shape == (3, 20)
assert algo.user_features_.shape == (3, 20)
preds = algo.predict_for_user(10, [3])
assert len(preds) == 1
assert preds.index[0] == 3
assert all(preds.notna())
def test_fsvd_predict_bad_item():
algo = svd.FunkSVD(20, iterations=20)
algo.fit(simple_df)
assert algo.global_bias_ == approx(simple_df.rating.mean())
assert algo.item_features_.shape == (3, 20)
assert algo.user_features_.shape == (3, 20)
preds = algo.predict_for_user(10, [4])
assert len(preds) == 1
assert preds.index[0] == 4
assert np.isnan(preds.loc[4])
def test_fsvd_predict_bad_item_clamp():
algo = svd.FunkSVD(20, iterations=20, range=(1, 5))
algo.fit(simple_df)
assert algo.global_bias_ == approx(simple_df.rating.mean())
assert algo.item_features_.shape == (3, 20)
assert algo.user_features_.shape == (3, 20)
preds = algo.predict_for_user(10, [4])
assert len(preds) == 1
assert preds.index[0] == 4
assert np.isnan(preds.loc[4])
def test_fsvd_predict_bad_user():
algo = svd.FunkSVD(20, iterations=20)
algo.fit(simple_df)
assert algo.global_bias_ == approx(simple_df.rating.mean())
assert algo.item_features_.shape == (3, 20)
assert algo.user_features_.shape == (3, 20)
preds = algo.predict_for_user(50, [3])
assert len(preds) == 1
assert preds.index[0] == 3
assert np.isnan(preds.loc[3])
@lktu.wantjit
@mark.slow
def test_fsvd_save_load():
ratings = lktu.ml_test.ratings
original = svd.FunkSVD(20, iterations=20)
original.fit(ratings)
assert original.global_bias_ == approx(ratings.rating.mean())
assert original.item_features_.shape == (ratings.item.nunique(), 20)
assert original.user_features_.shape == (ratings.user.nunique(), 20)
mod = pickle.dumps(original)
_log.info('serialized to %d bytes', len(mod))
algo = pickle.loads(mod)
assert algo.global_bias_ == original.global_bias_
assert np.all(algo.user_bias_ == original.user_bias_)
assert np.all(algo.item_bias_ == original.item_bias_)
assert np.all(algo.user_features_ == original.user_features_)
assert np.all(algo.item_features_ == original.item_features_)
assert np.all(algo.item_index_ == original.item_index_)
assert np.all(algo.user_index_ == original.user_index_)
@lktu.wantjit
@mark.slow
def test_fsvd_train_binary():
ratings = lktu.ml_test.ratings.drop(columns=['rating', 'timestamp'])
original = svd.FunkSVD(20, iterations=20, bias=False)
original.fit(ratings)
assert original.global_bias_ == 0
assert original.item_features_.shape == (ratings.item.nunique(), 20)
assert original.user_features_.shape == (ratings.user.nunique(), 20)
@lktu.wantjit
@mark.slow
def test_fsvd_known_preds():
algo = svd.FunkSVD(15, iterations=125, lrate=0.001)
_log.info('training %s on ml data', algo)
algo.fit(lktu.ml_test.ratings)
dir = Path(__file__).parent
pred_file = dir / 'funksvd-preds.csv'
_log.info('reading known predictions from %s', pred_file)
known_preds = pd.read_csv(str(pred_file))
pairs = known_preds.loc[:, ['user', 'item']]
preds = algo.predict(pairs)
known_preds.rename(columns={'prediction': 'expected'}, inplace=True)
merged = known_preds.assign(prediction=preds)
merged['error'] = merged.expected - merged.prediction
assert not any(merged.prediction.isna() & merged.expected.notna())
err = merged.error
err = err[err.notna()]
try:
assert all(err.abs() < 0.01)
except AssertionError as e:
bad = merged[merged.error.notna() & (merged.error.abs() >= 0.01)]
_log.error('erroneous predictions:\n%s', bad)
raise e
@lktu.wantjit
@mark.slow
@mark.eval
@mark.skipif(not lktu.ml100k.available, reason='ML100K data not present')
def test_fsvd_batch_accuracy():
from lenskit.algorithms import basic
import lenskit.crossfold as xf
from lenskit import batch
import lenskit.metrics.predict as pm
ratings = lktu.ml100k.ratings
svd_algo = svd.FunkSVD(25, 125, damping=10)
algo = basic.Fallback(svd_algo, basic.Bias(damping=10))
def eval(train, test):
_log.info('running training')
algo.fit(train)
_log.info('testing %d users', test.user.nunique())
return batch.predict(algo, test)
folds = xf.partition_users(ratings, 5, xf.SampleFrac(0.2))
preds = pd.concat(eval(train, test) for (train, test) in folds)
mae = pm.mae(preds.prediction, preds.rating)
assert mae == approx(0.74, abs=0.025)
user_rmse = preds.groupby('user').apply(lambda df: pm.rmse(df.prediction, df.rating))
assert user_rmse.mean() == approx(0.92, abs=0.05)