disco 0.2.6 → 0.2.7
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +4 -4
- data/CHANGELOG.md +4 -0
- data/README.md +32 -13
- data/lib/disco/recommender.rb +30 -14
- data/lib/disco/version.rb +1 -1
- metadata +3 -3
checksums.yaml
CHANGED
@@ -1,7 +1,7 @@
|
|
1
1
|
---
|
2
2
|
SHA256:
|
3
|
-
metadata.gz:
|
4
|
-
data.tar.gz:
|
3
|
+
metadata.gz: 5f400f07839587b574ddcfa4c88335bfe20fcd876164b943e8094a35c3c1cfef
|
4
|
+
data.tar.gz: e2426b283146837d14be154ff0e67eb2505fd6587958b39212bf2dfe3bfccd80
|
5
5
|
SHA512:
|
6
|
-
metadata.gz:
|
7
|
-
data.tar.gz:
|
6
|
+
metadata.gz: 2be9f24184036ec5b093de55640aebb60887ac59c566f37698fcba7a18daa15cf586566708def0060f80fc0747a50447538cf42fdf36024ae19ddac0de8b415c
|
7
|
+
data.tar.gz: 4682a5524a8cad4a247ec53f99c78e317d56ee55433bb2ad7806af4f2a9854bc016fd23564003f009dc69d0fdcf81949dc88c64d3cbe824a8e76fc5cae8abc7d
|
data/CHANGELOG.md
CHANGED
data/README.md
CHANGED
@@ -35,17 +35,15 @@ recommender.fit([
|
|
35
35
|
|
36
36
|
> IDs can be integers, strings, or any other data type
|
37
37
|
|
38
|
-
If users don’t rate items directly (for instance, they’re purchasing items or reading posts), this is known as implicit feedback. Leave out the rating
|
38
|
+
If users don’t rate items directly (for instance, they’re purchasing items or reading posts), this is known as implicit feedback. Leave out the rating.
|
39
39
|
|
40
40
|
```ruby
|
41
41
|
recommender.fit([
|
42
|
-
{user_id: 1, item_id: 1
|
43
|
-
{user_id: 2, item_id: 1
|
42
|
+
{user_id: 1, item_id: 1},
|
43
|
+
{user_id: 2, item_id: 1}
|
44
44
|
])
|
45
45
|
```
|
46
46
|
|
47
|
-
> Use `value` instead of `rating` for implicit feedback
|
48
|
-
|
49
47
|
Get user-based recommendations - “users like you also liked”
|
50
48
|
|
51
49
|
```ruby
|
@@ -106,11 +104,10 @@ views = Ahoy::Event.
|
|
106
104
|
count
|
107
105
|
|
108
106
|
data =
|
109
|
-
views.map do |(user_id, post_id),
|
107
|
+
views.map do |(user_id, post_id), _|
|
110
108
|
{
|
111
109
|
user_id: user_id,
|
112
|
-
item_id: post_id
|
113
|
-
value: count
|
110
|
+
item_id: post_id
|
114
111
|
}
|
115
112
|
end
|
116
113
|
```
|
@@ -201,7 +198,7 @@ bin = File.binread("recommender.bin")
|
|
201
198
|
recommender = Marshal.load(bin)
|
202
199
|
```
|
203
200
|
|
204
|
-
Alternatively, you can store only the factors and use a library like [Neighbor](https://github.com/ankane/neighbor)
|
201
|
+
Alternatively, you can store only the factors and use a library like [Neighbor](https://github.com/ankane/neighbor). See the [examples](https://github.com/ankane/neighbor/tree/master/examples).
|
205
202
|
|
206
203
|
## Algorithms
|
207
204
|
|
@@ -282,22 +279,22 @@ gem 'faiss'
|
|
282
279
|
Speed up the `user_recs` method with:
|
283
280
|
|
284
281
|
```ruby
|
285
|
-
|
282
|
+
recommender.optimize_user_recs
|
286
283
|
```
|
287
284
|
|
288
285
|
Speed up the `item_recs` method with:
|
289
286
|
|
290
287
|
```ruby
|
291
|
-
|
288
|
+
recommender.optimize_item_recs
|
292
289
|
```
|
293
290
|
|
294
291
|
Speed up the `similar_users` method with:
|
295
292
|
|
296
293
|
```ruby
|
297
|
-
|
294
|
+
recommender.optimize_similar_users
|
298
295
|
```
|
299
296
|
|
300
|
-
This should be called after fitting or loading the
|
297
|
+
This should be called after fitting or loading the recommender.
|
301
298
|
|
302
299
|
## Reference
|
303
300
|
|
@@ -336,6 +333,28 @@ Thanks to:
|
|
336
333
|
- [Implicit](https://github.com/benfred/implicit/) for serving as an initial reference for user and item similarity
|
337
334
|
- [@dasch](https://github.com/dasch) for the gem name
|
338
335
|
|
336
|
+
## Upgrading
|
337
|
+
|
338
|
+
### 0.2.7
|
339
|
+
|
340
|
+
There’s now a warning when passing `:value` with implicit feedback, as this has no effect on recommendations and can be removed. Earlier versions of the library incorrectly stated this was used.
|
341
|
+
|
342
|
+
```ruby
|
343
|
+
recommender.fit([
|
344
|
+
{user_id: 1, item_id: 1, value: 1},
|
345
|
+
{user_id: 2, item_id: 1, value: 3}
|
346
|
+
])
|
347
|
+
```
|
348
|
+
|
349
|
+
to:
|
350
|
+
|
351
|
+
```ruby
|
352
|
+
recommender.fit([
|
353
|
+
{user_id: 1, item_id: 1},
|
354
|
+
{user_id: 2, item_id: 1}
|
355
|
+
])
|
356
|
+
```
|
357
|
+
|
339
358
|
## History
|
340
359
|
|
341
360
|
View the [changelog](https://github.com/ankane/disco/blob/master/CHANGELOG.md)
|
data/lib/disco/recommender.rb
CHANGED
@@ -22,6 +22,10 @@ module Disco
|
|
22
22
|
# but may be confusing if they are all missing and later ones aren't
|
23
23
|
@implicit = !train_set.any? { |v| v[:rating] }
|
24
24
|
|
25
|
+
if @implicit && train_set.any? { |v| v[:value] }
|
26
|
+
warn "[disco] WARNING: Passing `:value` with implicit feedback has no effect on recommendations and can be removed. Earlier versions of the library incorrectly stated this was used."
|
27
|
+
end
|
28
|
+
|
25
29
|
# TODO improve performance
|
26
30
|
# (catch exception instead of checking ahead of time)
|
27
31
|
unless @implicit
|
@@ -34,7 +38,6 @@ module Disco
|
|
34
38
|
|
35
39
|
@rated = Hash.new { |hash, key| hash[key] = {} }
|
36
40
|
input = []
|
37
|
-
value_key = @implicit ? :value : :rating
|
38
41
|
train_set.each do |v|
|
39
42
|
# update maps and build matrix in single pass
|
40
43
|
u = (@user_map[v[:user_id]] ||= @user_map.size)
|
@@ -42,7 +45,7 @@ module Disco
|
|
42
45
|
@rated[u][i] = true
|
43
46
|
|
44
47
|
# explicit will always have a value due to check_ratings
|
45
|
-
input << [u, i,
|
48
|
+
input << [u, i, @implicit ? 1 : v[:rating]]
|
46
49
|
end
|
47
50
|
@rated.default = nil
|
48
51
|
|
@@ -61,7 +64,7 @@ module Disco
|
|
61
64
|
train_set.each do |v|
|
62
65
|
i = @item_map[v[:item_id]]
|
63
66
|
@item_count[i] += 1
|
64
|
-
@item_sum[i] += (v[
|
67
|
+
@item_sum[i] += (@implicit ? 1 : v[:rating])
|
65
68
|
end
|
66
69
|
end
|
67
70
|
|
@@ -76,7 +79,7 @@ module Disco
|
|
76
79
|
u ||= -1
|
77
80
|
i ||= -1
|
78
81
|
|
79
|
-
eval_set << [u, i,
|
82
|
+
eval_set << [u, i, @implicit ? 1 : v[:rating]]
|
80
83
|
end
|
81
84
|
end
|
82
85
|
|
@@ -138,8 +141,7 @@ module Disco
|
|
138
141
|
predictions, ids = @user_recs_index.search(@user_factors[u, true].expand_dims(0), count + rated.size).map { |v| v[0, true] }
|
139
142
|
else
|
140
143
|
predictions = @item_factors.inner(@user_factors[u, true])
|
141
|
-
|
142
|
-
indexes = predictions.sort_index.reverse
|
144
|
+
indexes = predictions.sort_index.reverse # reverse just creates view
|
143
145
|
indexes = indexes[0...[count + rated.size, indexes.size].min] if count
|
144
146
|
predictions = predictions[indexes]
|
145
147
|
ids = indexes
|
@@ -179,19 +181,32 @@ module Disco
|
|
179
181
|
raise "top_items not computed" unless @top_items
|
180
182
|
|
181
183
|
if @implicit
|
182
|
-
scores = @item_count
|
184
|
+
scores = Numo::UInt64.cast(@item_count)
|
183
185
|
else
|
184
186
|
require "wilson_score"
|
185
187
|
|
186
188
|
range = @min_rating..@max_rating
|
187
|
-
scores = @item_sum.zip(@item_count).map { |s, c| WilsonScore.rating_lower_bound(s / c, c, range) }
|
189
|
+
scores = Numo::DFloat.cast(@item_sum.zip(@item_count).map { |s, c| WilsonScore.rating_lower_bound(s / c, c, range) })
|
190
|
+
|
191
|
+
# TODO uncomment in 0.3.0
|
192
|
+
# wilson score with continuity correction
|
193
|
+
# https://en.wikipedia.org/wiki/Binomial_proportion_confidence_interval#Wilson_score_interval_with_continuity_correction
|
194
|
+
# z = 1.96 # 95% confidence
|
195
|
+
# range = @max_rating - @min_rating
|
196
|
+
# n = Numo::DFloat.cast(@item_count)
|
197
|
+
# phat = (Numo::DFloat.cast(@item_sum) - (@min_rating * n)) / range / n
|
198
|
+
# phat = (phat - (1 / 2 * n)).clip(0, 100) # continuity correction
|
199
|
+
# scores = (phat + z**2 / (2 * n) - z * Numo::DFloat::Math.sqrt((phat * (1 - phat) + z**2 / (4 * n)) / n)) / (1 + z**2 / n)
|
200
|
+
# scores = scores * range + @min_rating
|
188
201
|
end
|
189
202
|
|
190
|
-
|
191
|
-
|
192
|
-
|
193
|
-
|
194
|
-
|
203
|
+
indexes = scores.sort_index.reverse
|
204
|
+
indexes = indexes[0...[count, indexes.size].min] if count
|
205
|
+
scores = scores[indexes]
|
206
|
+
|
207
|
+
keys = @item_map.keys
|
208
|
+
indexes.size.times.map do |i|
|
209
|
+
{item_id: keys[indexes[i]], score: scores[i]}
|
195
210
|
end
|
196
211
|
end
|
197
212
|
|
@@ -255,8 +270,9 @@ module Disco
|
|
255
270
|
# inner product is cosine similarity with normalized vectors
|
256
271
|
# https://github.com/facebookresearch/faiss/issues/95
|
257
272
|
#
|
258
|
-
# TODO use non-exact index
|
273
|
+
# TODO use non-exact index in 0.3.0
|
259
274
|
# https://github.com/facebookresearch/faiss/wiki/Faiss-indexes
|
275
|
+
# index = Faiss::IndexHNSWFlat.new(factors.shape[1], 32, :inner_product)
|
260
276
|
index = Faiss::IndexFlatIP.new(factors.shape[1])
|
261
277
|
|
262
278
|
# ids are from 0...total
|
data/lib/disco/version.rb
CHANGED
metadata
CHANGED
@@ -1,14 +1,14 @@
|
|
1
1
|
--- !ruby/object:Gem::Specification
|
2
2
|
name: disco
|
3
3
|
version: !ruby/object:Gem::Version
|
4
|
-
version: 0.2.
|
4
|
+
version: 0.2.7
|
5
5
|
platform: ruby
|
6
6
|
authors:
|
7
7
|
- Andrew Kane
|
8
8
|
autorequire:
|
9
9
|
bindir: bin
|
10
10
|
cert_chain: []
|
11
|
-
date: 2021-
|
11
|
+
date: 2021-08-06 00:00:00.000000000 Z
|
12
12
|
dependencies:
|
13
13
|
- !ruby/object:Gem::Dependency
|
14
14
|
name: libmf
|
@@ -76,7 +76,7 @@ required_rubygems_version: !ruby/object:Gem::Requirement
|
|
76
76
|
- !ruby/object:Gem::Version
|
77
77
|
version: '0'
|
78
78
|
requirements: []
|
79
|
-
rubygems_version: 3.2.
|
79
|
+
rubygems_version: 3.2.22
|
80
80
|
signing_key:
|
81
81
|
specification_version: 4
|
82
82
|
summary: Recommendations for Ruby and Rails using collaborative filtering
|