-
Notifications
You must be signed in to change notification settings - Fork 26
/
reclist.py
287 lines (253 loc) · 10.9 KB
/
reclist.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
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
import collections
from reclist.abstractions import RecList, rec_test
from typing import List
import random
class CoveoCartRecList(RecList):
@rec_test(test_type='stats')
def basic_stats(self):
"""
Basic statistics on training, test and prediction data
"""
from reclist.metrics.standard_metrics import statistics
return statistics(self._x_train,
self._y_train,
self._x_test,
self._y_test,
self._y_preds)
@rec_test(test_type='price_homogeneity')
def price_test(self):
"""
Measures the absolute log ratio of ground truth and prediction price
"""
from reclist.metrics.price_homogeneity import price_homogeneity_test
return price_homogeneity_test(y_test=self.sku_only(self._y_test),
y_preds=self.sku_only(self._y_preds),
product_data=self.product_data,
price_sel_fn=lambda x: float(x['price_bucket'])
if x['price_bucket']
else None
)
@rec_test(test_type='Coverage@10')
def coverage_at_k(self):
"""
Coverage is the proportion of all possible products which the RS
recommends based on a set of sessions
"""
from reclist.metrics.standard_metrics import coverage_at_k
return coverage_at_k(self.sku_only(self._y_preds),
self.product_data,
k=10)
@rec_test(test_type='HR@10')
def hit_rate_at_k(self):
"""
Compute the rate in which the top-k predictions contain the item to be predicted
"""
from reclist.metrics.standard_metrics import hit_rate_at_k
return hit_rate_at_k(self.sku_only(self._y_preds),
self.sku_only(self._y_test),
k=10)
@rec_test(test_type='hits_distribution')
def hits_distribution(self):
"""
Compute the distribution of hit-rate across product frequency in training data
"""
from reclist.metrics.hits import hits_distribution
return hits_distribution(self.sku_only(self._x_train),
self.sku_only(self._x_test),
self.sku_only(self._y_test),
self.sku_only(self._y_preds),
k=10,
debug=True)
@rec_test(test_type='distance_to_query')
def dist_to_query(self):
"""
Compute the distribution of distance from query to label and query to prediction
"""
from reclist.metrics.distance_metrics import distance_to_query
return distance_to_query(self.rec_model,
self.sku_only(self._x_test),
self.sku_only(self._y_test),
self.sku_only(self._y_preds), k=10, bins=25, debug=True)
def sku_only(self, l: List[List]):
return [[e['product_sku'] for e in s] for s in l]
class SpotifySessionRecList(RecList):
@rec_test(test_type='basic_stats')
def basic_stats(self):
"""
Basic statistics on training, test and prediction data for Next Event Prediction
"""
from reclist.metrics.standard_metrics import statistics
return statistics(self._x_train,
self._y_train,
self._x_test,
self._y_test,
self._y_preds)
@rec_test(test_type='HR@10')
def hit_rate_at_k(self):
"""
Compute the rate at which the top-k predictions contain the item to be predicted
"""
from reclist.metrics.standard_metrics import hit_rate_at_k
return hit_rate_at_k(self.uri_only(self._y_preds),
self.uri_only(self._y_test),
k=10)
@rec_test(test_type='perturbation_test')
def perturbation_at_k(self):
"""
Compute average consistency in model predictions when inputs are perturbed
"""
from reclist.metrics.perturbation import session_perturbation_test
from collections import defaultdict
from functools import partial
# Step 1: Generate a map from artist uri to track uri
substitute_mapping = defaultdict(list)
for track_uri, row in self.product_data.items():
substitute_mapping[row['artist_uri']].append(track_uri)
# Step 2: define a custom perturbation function
def perturb(session, sub_map):
last_item = session[-1]
last_item_artist = self.product_data[last_item['track_uri']]['artist_uri']
substitutes = set(sub_map.get(last_item_artist,[])) - {last_item['track_uri']}
if substitutes:
similar_item = random.sample(substitutes, k=1)
new_session = session[:-1] + [{"track_uri": similar_item[0]}]
return new_session
return []
# Step 3: call test
return session_perturbation_test(self.rec_model,
self._x_test,
self._y_preds,
partial(perturb, sub_map=substitute_mapping),
self.uri_only,
k=10)
@rec_test(test_type='shuffle_session')
def perturbation_shuffle_at_k(self):
"""
Compute average consistency in model predictions when inputs are re-ordered
"""
from reclist.metrics.perturbation import session_perturbation_test
# Step 1: define a custom perturbation function
def perturb(session):
return random.sample(session, len(session))
# Step 2: call test
return session_perturbation_test(self.rec_model,
self._x_test,
self._y_preds,
perturb,
self.uri_only,
k=10)
@rec_test(test_type='hits_distribution_by_slice')
def hits_distribution_by_slice(self):
"""
Compute the distribution of hit-rate across various slices of data
"""
from reclist.metrics.hits import hits_distribution_by_slice
len_map = collections.defaultdict(list)
for idx, playlist in enumerate(self._x_test):
len_map[len(playlist)].append(idx)
slices = collections.defaultdict(list)
bins = [(x * 5, (x + 1) * 5) for x in range(max(len_map) // 5 + 1)]
for bin_min, bin_max in bins:
for i in range(bin_min + 1, bin_max + 1, 1):
slices[f'({bin_min}, {bin_max}]'].extend(len_map[i])
del len_map[i]
assert len(len_map) == 0
return hits_distribution_by_slice(slices,
self.uri_only(self._y_test),
self.uri_only(self._y_preds),
debug=True)
@rec_test(test_type='Coverage@10')
def coverage_at_k(self):
"""
Coverage is the proportion of all possible products which the RS
recommends based on a set of sessions
"""
from reclist.metrics.standard_metrics import coverage_at_k
return coverage_at_k(self.uri_only(self._y_preds),
self.product_data,
# this contains all the track URIs from train and test sets
k=10)
@rec_test(test_type='Popularity@10')
def popularity_bias_at_k(self):
"""
Compute average frequency of occurrence across recommended items in training data
"""
from reclist.metrics.standard_metrics import popularity_bias_at_k
return popularity_bias_at_k(self.uri_only(self._y_preds),
self.uri_only(self._x_train),
k=10)
@rec_test(test_type='MRR@10')
def mrr_at_k(self):
"""
MRR calculates the mean reciprocal of the rank at which the first
relevant item was retrieved
"""
from reclist.metrics.standard_metrics import mrr_at_k
return mrr_at_k(self.uri_only(self._y_preds),
self.uri_only(self._y_test))
def uri_only(self, playlists: List[dict]):
return [[track['track_uri'] for track in playlist] for playlist in playlists]
class MovieLensSimilarItemRecList(RecList):
@rec_test(test_type="stats")
def basic_stats(self):
"""
Basic statistics on training, test and prediction data
"""
from reclist.metrics.standard_metrics import statistics
return statistics(
self._x_train,
self._y_train,
self._x_test,
self._y_test,
self._y_preds
)
@rec_test(test_type='HR@10')
def hit_rate_at_k(self):
"""
Compute the rate at which the top-k predictions contain the movie to be predicted
"""
from reclist.metrics.standard_metrics import hit_rate_at_k
return hit_rate_at_k(
self.movie_only(self._y_preds),
self.movie_only(self._y_test),
k=10
)
@rec_test(test_type='Coverage@10')
def coverage_at_k(self):
"""
Coverage is the proportion of all possible movies which the RS
recommends based on a set of movies and their respective ratings
"""
from reclist.metrics.standard_metrics import coverage_at_k
return coverage_at_k(
self.movie_only(self._y_preds),
self.product_data,
k=10
)
@rec_test(test_type='hits_distribution')
def hits_distribution(self):
"""
Compute the distribution of hit-rate across movie frequency in training data
"""
from reclist.metrics.hits import hits_distribution
return hits_distribution(
self.movie_only(self._x_train),
self.movie_only(self._x_test),
self.movie_only(self._y_test),
self.movie_only(self._y_preds),
k=10,
debug=True
)
@rec_test(test_type="hits_distribution_by_rating")
def hits_distribution_by_rating(self):
"""
Compute the distribution of hit-rate across movie ratings in testing data
"""
from reclist.metrics.hits import hits_distribution_by_rating
return hits_distribution_by_rating(
self._y_test,
self._y_preds,
debug=True
)
def movie_only(self, movies):
return [[x["movieId"] for x in y] for y in movies]