File size: 9,376 Bytes
133a630
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
60c3ccb
 
 
133a630
 
 
 
 
 
 
 
 
 
 
 
 
daa3d2a
 
133a630
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
60c3ccb
 
 
133a630
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
60c3ccb
 
 
 
 
 
 
 
 
 
 
 
133a630
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
60c3ccb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
133a630
 
 
 
 
60c3ccb
 
 
 
133a630
 
 
 
 
 
 
 
 
 
 
 
60c3ccb
 
133a630
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import numpy as np
import pandas as pd
from utils.similarity import cosine_similarity, pearson_similarity, adjusted_cosine_similarity
from utils.helpers import build_user_item_matrix


class CollaborativeFiltering:
    def __init__(self, ratings_df):
        self.ratings = ratings_df
        self.matrix = build_user_item_matrix(ratings_df)
        self.user_item_matrix = self.matrix.values
        self.n_users, self.n_items = self.user_item_matrix.shape
        self.user_ids = self.matrix.index.values
        self.item_ids = self.matrix.columns.values

        self.user_means = np.nanmean(self.user_item_matrix, axis=1)
        self.global_mean = np.nanmean(self.user_item_matrix)

        self._svd_cache = None
        self._slope_one_dev = None

    def _get_user_index(self, user_id):
        indices = np.where(self.user_ids == user_id)[0]
        return indices[0] if len(indices) > 0 else None

    def _get_item_index(self, item_id):
        indices = np.where(self.item_ids == item_id)[0]
        return indices[0] if len(indices) > 0 else None

    def user_based_cf(self, user_id, n_recommendations=10, k=20):
        u_idx = self._get_user_index(user_id)
        if u_idx is None:
            return []

        matrix_filled = np.nan_to_num(self.user_item_matrix, nan=self.global_mean)
        sim_matrix = cosine_similarity(matrix_filled)
        user_sim = sim_matrix[u_idx]
        user_sim[u_idx] = 0

        user_ratings = self.user_item_matrix[u_idx]
        unseen = np.where(np.isnan(user_ratings))[0]
        if len(unseen) == 0:
            return []

        predictions = []
        for i_idx in unseen:
            similar_users = np.argsort(user_sim)[::-1][:k]
            valid = []
            for su in similar_users:
                if not np.isnan(self.user_item_matrix[su, i_idx]) and user_sim[su] > 0:
                    valid.append(su)
            if not valid:
                continue
            sim_vals = user_sim[valid]
            ratings_vals = self.user_item_matrix[valid, i_idx]
            pred = np.average(ratings_vals, weights=sim_vals)
            predictions.append((int(self.item_ids[i_idx]), float(pred)))

        predictions.sort(key=lambda x: x[1], reverse=True)
        return predictions[:n_recommendations]

    def item_based_cf(self, user_id, n_recommendations=10, k=15):
        u_idx = self._get_user_index(user_id)
        if u_idx is None:
            return []

        item_sim = adjusted_cosine_similarity(self.user_item_matrix)
        user_ratings = self.user_item_matrix[u_idx]
        unseen = np.where(np.isnan(user_ratings))[0]
        rated = np.where(~np.isnan(user_ratings))[0]

        if len(rated) == 0:
            return []

        predictions = []
        for i_idx in unseen:
            sim_to_rated = item_sim[i_idx, rated]
            best = np.argsort(sim_to_rated)[::-1][:k]
            valid = [(r, sim_to_rated[r]) for r in best if sim_to_rated[r] > 0 and r < len(rated)]
            if not valid:
                continue
            neighbor_indices = [rated[r[0]] for r in valid]
            sim_vals = [r[1] for r in valid]
            rating_vals = user_ratings[neighbor_indices]
            pred = np.average(rating_vals, weights=sim_vals)
            predictions.append((int(self.item_ids[i_idx]), float(pred)))

        predictions.sort(key=lambda x: x[1], reverse=True)
        return predictions[:n_recommendations]

    def train_svd_generator(self, n_factors=20, n_epochs=100, lr=0.01, reg=0.02):
        if self._svd_cache is not None:
            return

        matrix_imputed = self.user_item_matrix.copy()
        matrix_imputed = np.nan_to_num(matrix_imputed, nan=self.global_mean)

        n_u, n_i = matrix_imputed.shape
        np.random.seed(42)
        P = np.random.normal(0, 0.1, (n_u, n_factors))
        Q = np.random.normal(0, 0.1, (n_i, n_factors))
        bu = np.zeros(n_u)
        bi = np.zeros(n_i)

        observed = []
        for u in range(n_u):
            for i in range(n_i):
                if not np.isnan(self.user_item_matrix[u, i]):
                    observed.append((u, i))

        for epoch in range(n_epochs):
            np.random.shuffle(observed)
            for u, i in observed:
                r = self.user_item_matrix[u, i]
                pred = self.global_mean + bu[u] + bi[i] + np.dot(P[u], Q[i])
                err = r - pred
                bu[u] += lr * (err - reg * bu[u])
                bi[i] += lr * (err - reg * bi[i])
                P[u] += lr * (err * Q[i] - reg * P[u])
                Q[i] += lr * (err * P[u] - reg * Q[i])
            yield epoch + 1, n_epochs

        self._svd_cache = (P, Q, bu, bi)

    def svd(self, user_id, n_recommendations=10, n_factors=20, n_epochs=100, lr=0.01, reg=0.02):
        u_idx = self._get_user_index(user_id)
        if u_idx is None:
            return []

        for _ in self.train_svd_generator(n_factors, n_epochs, lr, reg):
            pass
        P, Q, bu, bi = self._svd_cache

        user_ratings = self.user_item_matrix[u_idx]
        unseen = np.where(np.isnan(user_ratings))[0]

        predictions = []
        for i_idx in unseen:
            pred = self.global_mean + bu[u_idx] + bi[i_idx] + np.dot(P[u_idx], Q[i_idx])
            predictions.append((int(self.item_ids[i_idx]), float(pred)))

        predictions.sort(key=lambda x: x[1], reverse=True)
        return predictions[:n_recommendations]

    def knn_cf(self, user_id, n_recommendations=10, k=10):
        u_idx = self._get_user_index(user_id)
        if u_idx is None:
            return []

        from sklearn.neighbors import NearestNeighbors
        matrix_imputed = np.nan_to_num(self.user_item_matrix, nan=self.global_mean)
        nn = NearestNeighbors(n_neighbors=min(k + 1, self.n_users), metric="cosine")
        nn.fit(matrix_imputed)
        distances, indices = nn.kneighbors(matrix_imputed[u_idx].reshape(1, -1))
        neighbor_indices = indices[0][1:]

        user_ratings = self.user_item_matrix[u_idx]
        unseen = np.where(np.isnan(user_ratings))[0]

        predictions = []
        for i_idx in unseen:
            neighbor_ratings = []
            neighbor_dists = []
            for ni in neighbor_indices:
                if not np.isnan(self.user_item_matrix[ni, i_idx]):
                    neighbor_ratings.append(self.user_item_matrix[ni, i_idx])
                    neighbor_dists.append(distances[0][list(indices[0]).index(ni)] + 1e-6)
            if not neighbor_ratings:
                continue
            weights = 1.0 / np.array(neighbor_dists)
            pred = np.average(neighbor_ratings, weights=weights)
            predictions.append((int(self.item_ids[i_idx]), float(pred)))

        predictions.sort(key=lambda x: x[1], reverse=True)
        return predictions[:n_recommendations]

    def compute_slope_one_dev_generator(self):
        if self._slope_one_dev is not None:
            return

        dev = np.zeros((self.n_items, self.n_items))
        cnt = np.zeros((self.n_items, self.n_items), dtype=int)

        for i in range(self.n_items):
            for j in range(self.n_items):
                if i == j:
                    continue
                diff_sum = 0.0
                count = 0
                for u in range(self.n_users):
                    vi = self.user_item_matrix[u, i]
                    vj = self.user_item_matrix[u, j]
                    if not np.isnan(vi) and not np.isnan(vj):
                        diff_sum += vi - vj
                        count += 1
                if count > 0:
                    dev[i, j] = diff_sum / count
                    cnt[i, j] = count
            yield i + 1, self.n_items

        self._slope_one_dev = (dev, cnt)

    def slope_one(self, user_id, n_recommendations=10):
        u_idx = self._get_user_index(user_id)
        if u_idx is None:
            return []

        for _ in self.compute_slope_one_dev_generator():
            pass
        dev, cnt = self._slope_one_dev

        user_ratings = self.user_item_matrix[u_idx]
        unseen = np.where(np.isnan(user_ratings))[0]
        rated = np.where(~np.isnan(user_ratings))[0]

        if len(rated) == 0:
            return []

        predictions = []
        for i_idx in unseen:
            numerator = 0.0
            denominator = 0.0
            for j_idx in rated:
                if cnt[i_idx, j_idx] > 0:
                    numerator += user_ratings[j_idx] + dev[i_idx, j_idx]
                    denominator += 1
            if denominator > 0:
                pred = numerator / denominator
                predictions.append((int(self.item_ids[i_idx]), float(pred)))

        predictions.sort(key=lambda x: x[1], reverse=True)
        return predictions[:n_recommendations]

    def recommend(self, method, user_id, n_recommendations=10, **kwargs):
        methods = {
            "user_based": self.user_based_cf,
            "item_based": self.item_based_cf,
            "svd": self.svd,
            "knn": self.knn_cf,
            "slope_one": self.slope_one,
        }
        func = methods.get(method)
        if func is None:
            raise ValueError(f"Unknown method: {method}")
        return func(user_id, n_recommendations=n_recommendations, **kwargs)

    def get_all_methods(self):
        return ["user_based", "item_based", "svd", "knn", "slope_one"]