-
Notifications
You must be signed in to change notification settings - Fork 58
/
test_als_explicit.py
157 lines (115 loc) · 4.66 KB
/
test_als_explicit.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
import logging
import pickle
from lenskit.algorithms import als
from lenskit import util
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_als_basic_build():
algo = als.BiasedMF(20, iterations=10, progress=util.no_progress)
algo.fit(simple_df)
assert algo.global_bias_ == approx(simple_df.rating.mean())
assert set(algo.user_index_) == set([10, 12, 13])
assert set(algo.item_index_) == set([1, 2, 3])
assert algo.user_features_.shape == (3, 20)
assert algo.item_features_.shape == (3, 20)
assert algo.n_features == 20
assert algo.n_users == 3
assert algo.n_items == 3
def test_als_no_bias():
algo = als.BiasedMF(20, iterations=10, bias=None)
algo.fit(simple_df)
assert algo.bias is None
assert algo.global_bias_ == 0
assert algo.item_bias_ is None
assert algo.user_bias_ is None
assert set(algo.user_index_) == set([10, 12, 13])
assert set(algo.item_index_) == set([1, 2, 3])
assert algo.user_features_.shape == (3, 20)
assert algo.item_features_.shape == (3, 20)
preds = algo.predict_for_user(10, [3])
assert len(preds) == 1
def test_als_predict_basic():
algo = als.BiasedMF(20, iterations=10)
algo.fit(simple_df)
assert algo.global_bias_ == approx(simple_df.rating.mean())
preds = algo.predict_for_user(10, [3])
assert len(preds) == 1
assert preds.index[0] == 3
assert preds.loc[3] >= -0.1
assert preds.loc[3] <= 5.1
def test_als_predict_bad_item():
algo = als.BiasedMF(20, iterations=10)
algo.fit(simple_df)
assert algo.global_bias_ == approx(simple_df.rating.mean())
preds = algo.predict_for_user(10, [4])
assert len(preds) == 1
assert preds.index[0] == 4
assert np.isnan(preds.loc[4])
def test_als_predict_bad_user():
algo = als.BiasedMF(20, iterations=10)
algo.fit(simple_df)
assert algo.global_bias_ == approx(simple_df.rating.mean())
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_als_train_large():
algo = als.BiasedMF(20, iterations=10)
ratings = lktu.ml_test.ratings
algo.fit(ratings)
assert algo.global_bias_ == approx(ratings.rating.mean())
assert algo.n_features == 20
assert algo.n_items == ratings.item.nunique()
assert algo.n_users == ratings.user.nunique()
icounts = ratings.groupby('item').rating.count()
isums = ratings.groupby('item').rating.sum()
is2 = isums - icounts * ratings.rating.mean()
imeans = is2 / (icounts + 5)
ibias = pd.Series(algo.item_bias_, index=algo.item_index_)
imeans, ibias = imeans.align(ibias)
assert ibias.values == approx(imeans.values)
# don't use wantjit, use this to do a non-JIT test
def test_als_save_load():
original = als.BiasedMF(20, iterations=5)
ratings = lktu.ml_test.ratings
original.fit(ratings)
assert original.global_bias_ == approx(ratings.rating.mean())
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_)
@mark.slow
@mark.eval
@mark.skipif(not lktu.ml100k.available, reason='ML100K data not present')
def test_als_batch_accuracy():
from lenskit.algorithms import basic
import lenskit.crossfold as xf
import lenskit.metrics.predict as pm
ratings = lktu.ml100k.ratings
svd_algo = als.BiasedMF(25, iterations=20, damping=5)
algo = basic.Fallback(svd_algo, basic.Bias(damping=5))
def eval(train, test):
_log.info('running training')
algo.fit(train)
_log.info('testing %d users', test.user.nunique())
return test.assign(prediction=algo.predict(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(0.73, abs=0.025)
user_rmse = preds.groupby('user').apply(lambda df: pm.rmse(df.prediction, df.rating))
assert user_rmse.mean() == approx(0.91, abs=0.05)