-
Notifications
You must be signed in to change notification settings - Fork 0
/
backend.py
415 lines (351 loc) · 17.9 KB
/
backend.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
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
from importlib.util import module_from_spec
import pandas as pd
import numpy as np
from sklearn.preprocessing import StandardScaler
from sklearn.cluster import KMeans
from sklearn.decomposition import PCA
from xgboost import XGBClassifier
from sklearn.preprocessing import MinMaxScaler
from sklearn.preprocessing import LabelEncoder
import keras
from tensorflow.keras import layers
import tensorflow as tf
import NN_Recommender
import pickle
import os
models = ("Course Similarity",
"User Profile",
"Clustering",
"Clustering with PCA",
"Neural Network",
"Classification with Embedding Features"
)
XGB_param = {'booster' : 'gbtree',
'eta':0.3,
'gamma': 0.1,
'max_depth': 10,
'subsample':1,
'scale_pos_weight':0.1
}
# ------- Functions ------
def load_ratings():
return pd.read_csv("ratings.csv")
def load_course_sims():
return pd.read_csv("sim.csv")
def load_courses():
df = pd.read_csv("course_processed.csv")
df['TITLE'] = df['TITLE'].str.title()
return df
def load_bow():
return pd.read_csv("courses_bows.csv")
def load_course_genre():
return pd.read_csv('course_genre.csv')
def load_users_profiles() :
return pd.read_csv('profile_genre.csv')
def load_nn_model():
nn_model = tf.keras.models.load_model('nn_trainedmodel')
return nn_model
def add_new_ratings(new_courses):
res_dict = {}
if len(new_courses) > 0:
# Create a new user id, max id + 1
ratings_df = load_ratings()
new_id = ratings_df['user'].max() + 1
users = [new_id] * len(new_courses)
ratings = [3.0] * len(new_courses)
res_dict['user'] = users
res_dict['item'] = new_courses
res_dict['rating'] = ratings
new_df = pd.DataFrame(res_dict)
updated_ratings = pd.concat([ratings_df, new_df])
updated_ratings.to_csv("ratings.csv", index=False)
return new_id
# Create course id to index and index to id mappings
def get_doc_dicts():
bow_df = load_bow()
grouped_df = bow_df.groupby(['doc_index', 'doc_id']).max().reset_index(drop=False)
idx_id_dict = grouped_df[['doc_id']].to_dict()['doc_id']
id_idx_dict = {v: k for k, v in idx_id_dict.items()}
del grouped_df
return idx_id_dict, id_idx_dict
def course_similarity_recommendations(idx_id_dict, id_idx_dict, enrolled_course_ids, sim_matrix):
all_courses = set(idx_id_dict.values())
unselected_course_ids = all_courses.difference(enrolled_course_ids)
# Create a dictionary to store your recommendation results
res = {}
# First find all enrolled courses for user
for enrolled_course in enrolled_course_ids:
for unselect_course in unselected_course_ids:
if enrolled_course in id_idx_dict and unselect_course in id_idx_dict:
idx1 = id_idx_dict[enrolled_course]
idx2 = id_idx_dict[unselect_course]
sim = sim_matrix[idx1][idx2]
if unselect_course not in res:
res[unselect_course] = sim
else:
if sim >= res[unselect_course]:
res[unselect_course] = sim
res = {k: v for k, v in sorted(res.items(), key=lambda item: item[1], reverse=True)}
return res
def generate_user_profile(user_id, course_genre_df, idx_id_dict):
all_courses = set(idx_id_dict.values())
ratings_df = load_ratings()
user_courses_df = ratings_df[ratings_df['user'] == user_id]
enrolled_course_ids = user_courses_df['item'].to_list()
user_profile_df = course_genre_df[course_genre_df['COURSE_ID'].isin(enrolled_course_ids)]
user_profile_vector = user_profile_df.iloc[:, 2:].to_numpy().sum(axis=0)*3
unkown_courses = all_courses.difference(enrolled_course_ids)
unknown_course_df = course_genre_df[course_genre_df['COURSE_ID'].isin(unkown_courses)]
return enrolled_course_ids, unknown_course_df, user_profile_vector
def generate_recomendation_scores(user_id, course_genre_df, idx_id_dict):
users = []
courses = []
scores = []
enrolled_course_ids, unknown_course_df, user_profile_vector = generate_user_profile(user_id,
course_genre_df,
idx_id_dict)
unknown_course_id = unknown_course_df['COURSE_ID'].values
recom_scores = np.dot(unknown_course_df.iloc[:, 2:].to_numpy(), user_profile_vector)
for i in range(len(unknown_course_id)) :
users.append(user_id)
courses.append(unknown_course_id[i])
scores.append(recom_scores[i])
return users, courses, scores
def combine_user_label(user_ids, labels):
labels_df = pd.DataFrame(labels)
cluster_df = pd.merge(user_ids, labels_df, left_index=True, right_index=True)
cluster_df.columns = ['user', 'cluster']
return cluster_df
def train_kmeans(n):
users_profiles_df = load_users_profiles()
scaler = StandardScaler()
user_profile_matrix = users_profiles_df.iloc[:, 1:].to_numpy()
scaled_matrix = scaler.fit_transform(user_profile_matrix)
kmeans_model = KMeans(n_clusters=n, max_iter=200, n_init=5, random_state=123)
kmeans_model.fit(scaled_matrix)
pickle.dump(scaler, open('kmeans\ScalerModel.pkl', 'wb'))
pickle.dump(kmeans_model, open('kmeans\KMeansModel.pkl', 'wb'))
def train_pca_kmeans(m, n):
users_profiles_df = load_users_profiles()
scaler = StandardScaler()
user_profile_matrix = users_profiles_df.iloc[:, 1:].to_numpy()
scaled_matrix = scaler.fit_transform(user_profile_matrix)
PCA_model = PCA(n_components=m, random_state=123)
PCA_model = PCA_model.fit(scaled_matrix)
PCA_vector = PCA_model.transform(scaled_matrix)
kmeans_model = KMeans(n_clusters=n, max_iter=200, n_init=5, random_state=123)
kmeans_model.fit(PCA_vector)
pickle.dump(scaler, open('PCA_kmeans\ScalerModel.pkl', 'wb'))
pickle.dump(PCA_model, open('PCA_kmeans\PCA_KMeans.pkl', 'wb'))
pickle.dump(kmeans_model, open('PCA_kmeans\KMeansModel.pkl', 'wb'))
def user_item_idx_id(data):
encoded_data = data.copy()
# Mapping user ids to indices
user_list = encoded_data["user"].unique().tolist()
user_id2idx_dict = {x: i for i, x in enumerate(user_list)}
user_idx2id_dict = {i: x for i, x in enumerate(user_list)}
# Mapping course ids to indices
course_list = encoded_data["item"].unique().tolist()
course_id2idx_dict = {x: i for i, x in enumerate(course_list)}
course_idx2id_dict = {i: x for i, x in enumerate(course_list)}
return user_id2idx_dict, user_idx2id_dict, course_id2idx_dict, course_idx2id_dict
def train_nn_recom(epochs=4):
rating_df = load_ratings()
num_users = len(rating_df['user'].unique())
num_items = len(rating_df['item'].unique())
user_id2idx_dict, user_idx2id_dict, course_id2idx_dict, course_idx2id_dict = user_item_idx_id(rating_df)
encoded_data = rating_df.copy()
# Convert original user ids to idx
encoded_data["user"] = encoded_data["user"].map(user_id2idx_dict)
# Convert original course ids to idx
encoded_data["item"] = encoded_data["item"].map(course_id2idx_dict)
# Convert rating to int
encoded_data["rating"] = encoded_data["rating"].values.astype("int")
scaler = MinMaxScaler()
encoded_data['rating'] = scaler.fit_transform(encoded_data['rating'].values.reshape(-1, 1))
X = encoded_data[['user', 'item']].values
y = encoded_data['rating'].values
nn_model = NN_Recommender.Neural_Network_Recommender(num_users, num_items)
nn_model.compile(optimizer='adam', loss='mean_squared_error', metrics='RootMeanSquaredError')
nn_model.fit(X, y, epochs=epochs)
print(nn_model.summary())
nn_model.save(filepath='nn_trainedmodel', save_format='tf', overwrite=True, include_optimizer=True)
print('Model Training Done Successfuly!')
return nn_model
def clf_w_embeding(epochs=4):
rating_df = load_ratings()
user_id2idx_dict, user_idx2id_dict, course_id2idx_dict, course_idx2id_dict = user_item_idx_id(rating_df)
nn_model = train_nn_recom(epochs=epochs)
user_embd = nn_model.get_layer('user_embedding_layer').get_weights()[0]
item_embd = nn_model.get_layer('item_embedding_layer').get_weights()[0]
user_embd = pd.DataFrame(user_embd)
item_embd = pd.DataFrame(item_embd)
user_embd.index = user_embd.index.map(user_idx2id_dict)
user_embd = user_embd.reset_index().rename(columns={'index':'user'})
# save new user's embd profile
user_embd[user_embd.user==user_embd.user.max()].to_csv('NewUser_embd.csv', index=False)
item_embd.index = item_embd.index.map(course_idx2id_dict)
item_embd = item_embd.reset_index().rename(columns={'index':'item'})
item_embd.to_csv('item_embd.csv', index=False)
embd_df = pd.merge(rating_df, user_embd, how='left', on='user')
embd_df = embd_df.merge(item_embd, how='left', on='item')
user_f = [f'{i}_x' for i in range(16)]
item_f = [f'{j}_y' for j in range(16)]
embd_f = [f'f{z}' for z in range(16)]
embd_df[embd_f] = embd_df[user_f] + embd_df[item_f].values
embd_df.drop((user_f + item_f), axis=1, inplace=True)
y = embd_df['rating'].values
y = LabelEncoder().fit_transform(y)
X = embd_df.drop(['user', 'item', 'rating'], axis=1)
XGB_model = XGBClassifier(**XGB_param, seed=rs)
XGB_model.fit(X, y)
pickle.dump(XGB_model, open('XGBoost\XGBoostModel.pkl', 'wb'))
# Model training
def train(model_name, params):
if model_name == models[2]:
if 'Number_of_Clusters' in params :
n = params['Number_of_Clusters']
train_kmeans(n)
elif model_name == models[3]:
m = params['Number_of_Components']
n = params['Number_of_Clusters']
train_pca_kmeans(m, n)
elif model_name == models[4]:
n = params['epochs']
train_nn_recom(n)
elif model_name == models[5] :
n = params['epochs']
clf_w_embeding(n)
else :
pass
# Prediction
def predict(model_name, user_ids, params):
idx_id_dict, id_idx_dict = get_doc_dicts()
sim_matrix = load_course_sims().to_numpy()
course_genre_df = load_course_genre()
users = []
courses = []
scores = []
res_dict = {}
for user_id in user_ids:
# Course Similarity model
if model_name == models[0]:
sim_threshold = 0.6
if "sim_threshold" in params:
sim_threshold = params["sim_threshold"] / 100.0
ratings_df = load_ratings()
user_ratings = ratings_df[ratings_df['user'] == user_id]
enrolled_course_ids = user_ratings['item'].to_list()
res = course_similarity_recommendations(idx_id_dict, id_idx_dict, enrolled_course_ids, sim_matrix)
for key, score in res.items():
if score >= sim_threshold:
users.append(user_id)
courses.append(key)
scores.append(score)
# User profile model
elif model_name == models[1]:
idx_id_dict, _ = get_doc_dicts()
users, courses, scores = generate_recomendation_scores(user_id, course_genre_df, idx_id_dict)
elif model_name == models[2]:
ratings_df = load_ratings()
users_profiles_df = load_users_profiles()
users_ids = users_profiles_df.loc[:, users_profiles_df.columns == 'user']
scaler = pickle.load(open('kmeans\ScalerModel.pkl', 'rb'))
kmeans_model = pickle.load(open('kmeans\KMeansModel.pkl', 'rb'))
labels = kmeans_model.labels_
cluster_df = combine_user_label(users_ids, labels)
cluster_df = pd.merge(ratings_df, cluster_df, on='user', how='inner').drop('rating', axis=1)
cluster_item_df = cluster_df.groupby(['cluster', 'item']).size().reset_index().rename(columns={0:'enrollments'})
# Create new user's profile
enrolled_course_ids, _, user_profile_vector = generate_user_profile(user_id,
course_genre_df,
idx_id_dict)
scaled_vector = scaler.transform(user_profile_vector.reshape(-1, 14))
user_cluster = kmeans_model.predict(scaled_vector)[0]
cluster_item_df = cluster_item_df[cluster_item_df['cluster']==user_cluster]
for user in user_ids :
cluster_courses = cluster_item_df['item'].to_list()
unenrolled_courses = [course for course in cluster_courses if course not in enrolled_course_ids]
for course in unenrolled_courses :
enroll_count = cluster_item_df[cluster_item_df['item']==course]['enrollments'].values[0]
if enroll_count > 100 :
users.append(user)
courses.append(course)
scores.append(enroll_count)
elif model_name == models[3]:
ratings_df = load_ratings()
users_profiles_df = load_users_profiles()
users_ids = users_profiles_df.loc[:, users_profiles_df.columns == 'user']
scaler = pickle.load(open('PCA_kmeans\ScalerModel.pkl', 'rb'))
PCA_model = pickle.load(open('PCA_kmeans\PCA_KMeans.pkl', 'rb'))
kmeans_model = pickle.load(open('PCA_kmeans\KMeansModel.pkl', 'rb'))
labels = kmeans_model.labels_
cluster_df = combine_user_label(users_ids, labels)
cluster_df = pd.merge(ratings_df, cluster_df, on='user', how='inner').drop('rating', axis=1)
cluster_item_df = cluster_df.groupby(['cluster', 'item']).size().reset_index().rename(columns={0:'enrollments'})
# Create new user's profile
enrolled_course_ids, _, user_profile_vector = generate_user_profile(user_id,
course_genre_df,
idx_id_dict)
scaled_vector = scaler.transform(user_profile_vector.reshape(-1, 14))
PCA_vector = PCA_model.transform(scaled_vector)
user_cluster = kmeans_model.predict(PCA_vector)[0]
cluster_item_df = cluster_item_df[cluster_item_df['cluster']==user_cluster]
for user in user_ids :
cluster_courses = cluster_item_df['item'].to_list()
unenrolled_courses = [course for course in cluster_courses if course not in enrolled_course_ids]
for course in unenrolled_courses :
enroll_count = cluster_item_df[cluster_item_df['item']==course]['enrollments'].values[0]
if enroll_count > 100 :
users.append(user)
courses.append(course)
scores.append(enroll_count)
elif model_name == models[4]:
nn_model = load_nn_model()
ratings_df = load_ratings()
new_user_df = ratings_df[ratings_df.user==user_id]
new_user_enrolled = new_user_df.item.to_list()
new_user_unenrolled = list(set(ratings_df.item.values).difference(new_user_enrolled))
new_user_df = pd.DataFrame({'user':user_id, 'item':new_user_unenrolled})
user_id2idx_dict, user_idx2id_dict, course_id2idx_dict, course_idx2id_dict = user_item_idx_id(ratings_df)
new_user_df['user'] = new_user_df['user'].map(user_id2idx_dict)
new_user_df['item'] = new_user_df['item'].map(course_id2idx_dict)
new_user_values = new_user_df.values
y_pred = nn_model.predict(new_user_values)
new_user_df['pred_score'] = y_pred
new_user_df.sort_values(by='pred_score', ascending=False, inplace=True)
new_user_df['user'] = new_user_df['user'].map(user_idx2id_dict)
new_user_df['item'] = new_user_df['item'].map(course_idx2id_dict)
users.append(user_id)
courses.extend(new_user_df['item'].to_list())
scores.extend(new_user_df['pred_score'].to_list())
elif model_name == models[5]:
rating_df = load_ratings()
XGB_model = pickle.load(open('XGBoost\XGBoostModel.pkl', 'rb'))
new_user_df = rating_df[rating_df.user==user_id]
new_user_enrolled = new_user_df.item.to_list()
new_user_unenrolled = list(set(rating_df.item.values).difference(new_user_enrolled))
new_user_df = pd.DataFrame({'user':user_id, 'item':new_user_unenrolled})
new_user_embd = pd.read_csv('NewUser_embd.csv')
item_embd = pd.read_csv('item_embd.csv')
unenrolled_item_embd = item_embd[item_embd.item.isin(new_user_unenrolled)]
new_user_embd_profile = unenrolled_item_embd.copy()
new_user_embd_profile.iloc[:,1:] = unenrolled_item_embd.iloc[:,1:] + new_user_embd.iloc[:,1:].values
new_X = new_user_embd_profile.iloc[:, 1:].values
pred_rate = XGB_model.predict_proba(new_X)[:, 1]
new_user_embd_profile['pred_rate'] = pred_rate
new_user_pred = new_user_embd_profile[['item','pred_rate']]
new_user_pred.sort_values(by='pred_rate', ascending=False, inplace=True)
users.append(user_id)
courses.extend(new_user_pred['item'].to_list())
scores.extend(new_user_pred['pred_rate'].to_list())
else :
pass
res_dict['USER'] = users*len(courses)
res_dict['COURSE_ID'] = courses
res_dict['SCORE'] = scores
res_df = pd.DataFrame(res_dict, columns=['USER', 'COURSE_ID', 'SCORE'])
res_df = res_df.sort_values(by='SCORE', ascending=False)
if 'top_courses' in params :
res_df = res_df[ : params['top_courses']]
return res_df