Kullanıcı \(i\) için film \(j\) tavsiye problemi için bir yaklaşım, eğitim verisinde filmlere verilen beğeni notlarını (ratings) bir \(i,j\) matrisinde tutmak, SVD ile bu matris üzerinde ayrıştırma yaparak hem kullanıcı hem film boyutlarında azaltılmış temsili vektörler elde etmek, ardından bu vektörleri mesela kullanıcı \(i\) vektörü ile tüm filmlerin azaltılmış temsildeki vektörleri ile çarpmak suretiyle o filmlere (çoğunlukla daha verilmemiş) notun ne olacağını tahmin etmeye uğraşmaktı.
Derin öğrenme, TensorFlow ile çözüm ne olabilir? Aslında üstteki ayrıştırma problemini TF ile çözmek mümkün, çünkü elde bir çarpım var, eğitim verisi var, çarpımın (tahmin) gerçek veriden olan farkı bir hata fonksiyonu olarak görülebilir, ve bu hatanın TF ile otomatik türevi alınarak minimizasyon noktası aranabilir.
Fakat tavsiye üretme mekanizmasını pür derin YSA bakışıyla çözmek belki daha iyi. Doğal dil işlemede (NLP) görülen kelime gömme (word embedding) kavramını bu problemde de kullanabiliriz. NLP problemlerinde elde bir doküman vardır, bir sözlükten gelen kelimeler bu belgelerde görülebilir. Kullanıcı / film, ya da müşteri / ürün bağlamında da sanki bir doküman vardır, ama bu dokümanda sadece tek bir beğeni için olan tek bir müşteri / ürün ikilisi vardır. NLP’de bir belge ” kedi paspasta oturuyor’‘, kimlik verisi üzerinden [200, 201, 565] olabiliyordu, kullanıcı / film için “belgede’’ mesela sadece [100, 4500] olur; kullanıcı 100 film 4500’i beğenmiş. Beğeni notunun nereye gittiğinin birazdan göreceğiz.
O zaman alttaki gibi bir ağ yapısı düşünülebilir; kullanıcı / film ikilisi kimlik verilerine gömme transformasyonu uygulanır, ve sonuç gömme vektörleri bu tabakada yanyana yer alırlar. Ardından gömme tabakasındaki tüm bilgi tam bağlanmış (fully connected) iki ayrı tabaka üzerinden tahmin üretmek için kullanılır, bu tahmin ise beğeni notunun ta kendisidir. Eldeki eğitim verisinde gerçek verilen not olacaktır, ve bir kayıp fonksiyonu üzerinden optimizasyon yapılır.
Meta Veri
Bu mimariye bazı ekler yapabiliriz. Mesela eğer müşteri, ürün hakkında “meta’’ bilgiler var ise, diyelim kullanıcıların yaşı, mesleği gibi, bu bilgiler de eğitim sırasında kullanılabilir. Kullanıcı meta bilgisi aynı kullanıcı için bir beğeniden diğerine değişmezler, o sebeple ürünler için ürünün hangi tarihte üretildiği, fiyatı, vs. gibi detaylar meta olabilir. Peki bu bilgileri nasıl DYSA’ya dahil ederiz?
Meta bilgisini direk gömme tabakasında sayısal kolonların hemen yanına koyabiliriz, mesela ürünün yapılış tarihi tek bir sayısal kolon olarak bu tabakada yer alabilir, yaş, vs. aynı şekilde. Mimari olarak, müşteri \(i\) ve ürün \(j\) için meta veri \(m_i,n_j\) olsun,
şeklinde olur.
Kodlamaya başlayabiliriz. Önce verileri okuyalım, eğitim test verisi yaratalım (DYSA paketi olarak TensorFlow üzerinde yazılmış onu biraz daha kolaylaştıran Keras paketi kullanıldı [1])
import pandas as pd
from keras.layers import Concatenate
from pathlib import Path
from zipfile import ZipFile
from keras.layers import Input, Embedding, Flatten, Dense, Dropout
from keras.models import Model
from sklearn.metrics import mean_squared_error, mean_absolute_error
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import scale
ML_100K_FOLDER = Path('ml-100k')
# Add 'item_id' to your column names list at the beginning
icols = ["item_id", "name", "date", "genre", "url"] # Added 'item_id'
icols += ["f" + str(x) for x in range(19)] # unused feature names (assuming 19 'genre' flags or similar)
rcols = ["user_id", "item_id", "rating", "timestamp"]
with ZipFile('ml-100k.zip', 'r') as z:
all_ratings = pd.read_csv(z.open('ml-100k/u.data'), sep='\t', names=rcols)
# Load items, explicitly setting 'item_id' as the index column
# The 'item_id' is the first column in u.item, so index_col=0
items = pd.read_csv(z.open('ml-100k/u.item'), sep='|', names=icols, encoding='latin-1', index_col=0) #
items.fillna(value="01-Jan-1997", inplace=True)
from sklearn.model_selection import train_test_split
ratings_train, ratings_test = train_test_split(
all_ratings, test_size=0.2, random_state=0)
print (ratings_train.shape)(80000, 4)Şimdi gömme tabakası kurulur, kullanıcı, film için bu 32’şer boyutlu olsun, yani yanyana 64 öğeli bir vektör.
user_id_input = Input(shape=[1])
item_id_input = Input(shape=[1])
meta_input = Input(shape=[1])
user_id_train = ratings_train['user_id'].values.reshape(-1, 1)
item_id_train = ratings_train['item_id'].values.reshape(-1, 1)
rating_train = ratings_train['rating']
user_id_test = ratings_test['user_id'].values.reshape(-1, 1)
item_id_test = ratings_test['item_id'].values.reshape(-1, 1)
rating_test = ratings_test['rating']
max_user_id = all_ratings['user_id'].max()
max_item_id = all_ratings['item_id'].max()
embedding_size = 32
user_embedding = Embedding(output_dim=embedding_size, input_dim=max_user_id + 1,
input_length=1)(user_id_input)
item_embedding = Embedding(output_dim=embedding_size, input_dim=max_item_id + 1,
input_length=1)(item_id_input)
user_vecs = Flatten()(user_embedding)
item_vecs = Flatten()(item_embedding)Şimdi bir meta kolonu ekleyelim, bunu sadece filmin üretim tarihi için yapacağız, fakat diğerleri için de yapılacak ek aynen bu şekilde olurdu,
parsed_dates = [int(film_date[-4:]) for film_date in items["date"].tolist()]
items['parsed_date'] = pd.Series(parsed_dates, index=items.index)
items['scaled_date'] = scale(items['parsed_date'].astype('float64'))
item_meta_train = items["scaled_date"].loc[ratings_train['item_id']].values.reshape(-1, 1)
item_meta_test = items["scaled_date"].loc[ratings_test['item_id']].values.reshape(-1, 1)
input_vecs = Concatenate()([user_vecs, item_vecs, meta_input])Şimdi tam bağlanmış tabakaları ekleyelim, arada bir de veri atma (dropout) tabakası olsun, bu tabaka içinden geçen verilerin bazılarını / rasgele bir şekilde yoksayıyor (eğitim döngüsünün o aşaması yapılmıyor yani), bu tekniğin aşırı uygunluk (overfitting) problemleri için faydalı olduğu keşfedildi.
Tabakalar hazırlanınca eğitime başlayabiliriz,
x = Dense(64, activation='relu')(input_vecs)
x = Dropout(0.5)(x)
x = Dense(32, activation='relu')(x)
y = Dense(1)(x)
model = Model(inputs=[user_id_input, item_id_input, meta_input], outputs=y)
model.compile(optimizer='adam', loss='mae')
# Pass NumPy arrays for prediction
initial_train_preds = model.predict([user_id_train, item_id_train, item_meta_train])
# Pass NumPy arrays for fitting
history = model.fit([user_id_train, item_id_train, item_meta_train], rating_train,
batch_size=64, epochs=15, validation_split=0.1,
shuffle=True)
test_preds = model.predict([user_id_test, item_id_test, item_meta_test])
print("Final test Loss: %0.3f" % mean_squared_error(test_preds, rating_test))
print("Final test Loss: %0.3f" % mean_absolute_error(test_preds, rating_test)) [1m 1/2500 [0m [37m-------------------- [0m [1m3:00 [0m 72ms/step [1m 69/2500 [0m [37m-------------------- [0m [1m1s [0m 736us/step [1m 144/2500 [0m [32m- [0m [37m------------------- [0m [1m1s [0m 705us/step [1m 239/2500 [0m [32m- [0m [37m------------------- [0m [1m1s [0m 635us/step [1m 337/2500 [0m [32m-- [0m [37m------------------ [0m [1m1s [0m 599us/step [1m 435/2500 [0m [32m--- [0m [37m----------------- [0m [1m1s [0m 579us/step [1m 529/2500 [0m [32m---- [0m [37m---------------- [0m [1m1s [0m 571us/step [1m 628/2500 [0m [32m----- [0m [37m--------------- [0m [1m1s [0m 562us/step [1m 722/2500 [0m [32m----- [0m [37m--------------- [0m [1m0s [0m 558us/step [1m 821/2500 [0m [32m------ [0m [37m-------------- [0m [1m0s [0m 552us/step [1m 920/2500 [0m [32m------- [0m [37m------------- [0m [1m0s [0m 547us/step [1m1018/2500 [0m [32m-------- [0m [37m------------ [0m [1m0s [0m 544us/step [1m1108/2500 [0m [32m-------- [0m [37m------------ [0m [1m0s [0m 545us/step [1m1204/2500 [0m [32m--------- [0m [37m----------- [0m [1m0s [0m 543us/step [1m1298/2500 [0m [32m---------- [0m [37m---------- [0m [1m0s [0m 543us/step [1m1390/2500 [0m [32m----------- [0m [37m--------- [0m [1m0s [0m 543us/step [1m1488/2500 [0m [32m----------- [0m [37m--------- [0m [1m0s [0m 541us/step [1m1579/2500 [0m [32m------------ [0m [37m-------- [0m [1m0s [0m 542us/step [1m1670/2500 [0m [32m------------- [0m [37m------- [0m [1m0s [0m 542us/step [1m1763/2500 [0m [32m-------------- [0m [37m------ [0m [1m0s [0m 542us/step [1m1855/2500 [0m [32m-------------- [0m [37m------ [0m [1m0s [0m 542us/step [1m1946/2500 [0m [32m--------------- [0m [37m----- [0m [1m0s [0m 543us/step [1m2040/2500 [0m [32m---------------- [0m [37m---- [0m [1m0s [0m 542us/step [1m2133/2500 [0m [32m----------------- [0m [37m--- [0m [1m0s [0m 542us/step [1m2227/2500 [0m [32m----------------- [0m [37m--- [0m [1m0s [0m 542us/step [1m2323/2500 [0m [32m------------------ [0m [37m-- [0m [1m0s [0m 541us/step [1m2423/2500 [0m [32m------------------- [0m [37m- [0m [1m0s [0m 540us/step [1m2500/2500 [0m [32m-------------------- [0m [37m [0m [1m1s [0m 545us/step
Epoch 1/15
[1m 1/1125 [0m [37m-------------------- [0m [1m14:07 [0m 754ms/step - loss: 3.7447 [1m 37/1125 [0m [37m-------------------- [0m [1m1s [0m 1ms/step - loss: 3.3329 [1m 77/1125 [0m [32m- [0m [37m------------------- [0m [1m1s [0m 1ms/step - loss: 3.0075 [1m 121/1125 [0m [32m-- [0m [37m------------------ [0m [1m1s [0m 1ms/step - loss: 2.6643 [1m 162/1125 [0m [32m-- [0m [37m------------------ [0m [1m1s [0m 1ms/step - loss: 2.4315 [1m 203/1125 [0m [32m--- [0m [37m----------------- [0m [1m1s [0m 1ms/step - loss: 2.2575 [1m 245/1125 [0m [32m---- [0m [37m---------------- [0m [1m1s [0m 1ms/step - loss: 2.1194 [1m 288/1125 [0m [32m----- [0m [37m--------------- [0m [1m1s [0m 1ms/step - loss: 2.0072 [1m 331/1125 [0m [32m----- [0m [37m--------------- [0m [1m0s [0m 1ms/step - loss: 1.9152 [1m 374/1125 [0m [32m------ [0m [37m-------------- [0m [1m0s [0m 1ms/step - loss: 1.8384 [1m 420/1125 [0m [32m------- [0m [37m------------- [0m [1m0s [0m 1ms/step - loss: 1.7694 [1m 461/1125 [0m [32m-------- [0m [37m------------ [0m [1m0s [0m 1ms/step - loss: 1.7167 [1m 504/1125 [0m [32m-------- [0m [37m------------ [0m [1m0s [0m 1ms/step - loss: 1.6682 [1m 545/1125 [0m [32m--------- [0m [37m----------- [0m [1m0s [0m 1ms/step - loss: 1.6275 [1m 592/1125 [0m [32m---------- [0m [37m---------- [0m [1m0s [0m 1ms/step - loss: 1.5861 [1m 635/1125 [0m [32m----------- [0m [37m--------- [0m [1m0s [0m 1ms/step - loss: 1.5524 [1m 674/1125 [0m [32m----------- [0m [37m--------- [0m [1m0s [0m 1ms/step - loss: 1.5246 [1m 715/1125 [0m [32m------------ [0m [37m-------- [0m [1m0s [0m 1ms/step - loss: 1.4979 [1m 755/1125 [0m [32m------------- [0m [37m------- [0m [1m0s [0m 1ms/step - loss: 1.4740 [1m 797/1125 [0m [32m-------------- [0m [37m------ [0m [1m0s [0m 1ms/step - loss: 1.4509 [1m 839/1125 [0m [32m-------------- [0m [37m------ [0m [1m0s [0m 1ms/step - loss: 1.4295 [1m 880/1125 [0m [32m--------------- [0m [37m----- [0m [1m0s [0m 1ms/step - loss: 1.4101 [1m 923/1125 [0m [32m---------------- [0m [37m---- [0m [1m0s [0m 1ms/step - loss: 1.3911 [1m 965/1125 [0m [32m----------------- [0m [37m--- [0m [1m0s [0m 1ms/step - loss: 1.3739 [1m1007/1125 [0m [32m----------------- [0m [37m--- [0m [1m0s [0m 1ms/step - loss: 1.3577 [1m1051/1125 [0m [32m------------------ [0m [37m-- [0m [1m0s [0m 1ms/step - loss: 1.3419 [1m1098/1125 [0m [32m------------------- [0m [37m- [0m [1m0s [0m 1ms/step - loss: 1.3260 [1m1125/1125 [0m [32m-------------------- [0m [37m [0m [1m2s [0m 2ms/step - loss: 1.3171 - val_loss: 0.7724
Epoch 2/15
[1m 1/1125 [0m [37m-------------------- [0m [1m20s [0m 19ms/step - loss: 0.7542 [1m 37/1125 [0m [37m-------------------- [0m [1m1s [0m 1ms/step - loss: 0.7812 [1m 75/1125 [0m [32m- [0m [37m------------------- [0m [1m1s [0m 1ms/step - loss: 0.7778 [1m 111/1125 [0m [32m- [0m [37m------------------- [0m [1m1s [0m 1ms/step - loss: 0.7761 [1m 150/1125 [0m [32m-- [0m [37m------------------ [0m [1m1s [0m 1ms/step - loss: 0.7736 [1m 189/1125 [0m [32m--- [0m [37m----------------- [0m [1m1s [0m 1ms/step - loss: 0.7712 [1m 231/1125 [0m [32m---- [0m [37m---------------- [0m [1m1s [0m 1ms/step - loss: 0.7702 [1m 274/1125 [0m [32m---- [0m [37m---------------- [0m [1m1s [0m 1ms/step - loss: 0.7696 [1m 315/1125 [0m [32m----- [0m [37m--------------- [0m [1m1s [0m 1ms/step - loss: 0.7693 [1m 356/1125 [0m [32m------ [0m [37m-------------- [0m [1m0s [0m 1ms/step - loss: 0.7691 [1m 396/1125 [0m [32m------- [0m [37m------------- [0m [1m0s [0m 1ms/step - loss: 0.7691 [1m 438/1125 [0m [32m------- [0m [37m------------- [0m [1m0s [0m 1ms/step - loss: 0.7691 [1m 480/1125 [0m [32m-------- [0m [37m------------ [0m [1m0s [0m 1ms/step - loss: 0.7691 [1m 522/1125 [0m [32m--------- [0m [37m----------- [0m [1m0s [0m 1ms/step - loss: 0.7690 [1m 562/1125 [0m [32m--------- [0m [37m----------- [0m [1m0s [0m 1ms/step - loss: 0.7689 [1m 599/1125 [0m [32m---------- [0m [37m---------- [0m [1m0s [0m 1ms/step - loss: 0.7688 [1m 639/1125 [0m [32m----------- [0m [37m--------- [0m [1m0s [0m 1ms/step - loss: 0.7686 [1m 682/1125 [0m [32m------------ [0m [37m-------- [0m [1m0s [0m 1ms/step - loss: 0.7684 [1m 726/1125 [0m [32m------------ [0m [37m-------- [0m [1m0s [0m 1ms/step - loss: 0.7682 [1m 771/1125 [0m [32m------------- [0m [37m------- [0m [1m0s [0m 1ms/step - loss: 0.7680 [1m 816/1125 [0m [32m-------------- [0m [37m------ [0m [1m0s [0m 1ms/step - loss: 0.7677 [1m 858/1125 [0m [32m--------------- [0m [37m----- [0m [1m0s [0m 1ms/step - loss: 0.7675 [1m 897/1125 [0m [32m--------------- [0m [37m----- [0m [1m0s [0m 1ms/step - loss: 0.7673 [1m 937/1125 [0m [32m---------------- [0m [37m---- [0m [1m0s [0m 1ms/step - loss: 0.7671 [1m 980/1125 [0m [32m----------------- [0m [37m--- [0m [1m0s [0m 1ms/step - loss: 0.7669 [1m1023/1125 [0m [32m------------------ [0m [37m-- [0m [1m0s [0m 1ms/step - loss: 0.7666 [1m1066/1125 [0m [32m------------------ [0m [37m-- [0m [1m0s [0m 1ms/step - loss: 0.7664 [1m1109/1125 [0m [32m------------------- [0m [37m- [0m [1m0s [0m 1ms/step - loss: 0.7662 [1m1125/1125 [0m [32m-------------------- [0m [37m [0m [1m2s [0m 1ms/step - loss: 0.7662 - val_loss: 0.7711
Epoch 3/15
[1m 1/1125 [0m [37m-------------------- [0m [1m23s [0m 21ms/step - loss: 0.8475 [1m 36/1125 [0m [37m-------------------- [0m [1m1s [0m 1ms/step - loss: 0.7436 [1m 73/1125 [0m [32m- [0m [37m------------------- [0m [1m1s [0m 1ms/step - loss: 0.7358 [1m 112/1125 [0m [32m- [0m [37m------------------- [0m [1m1s [0m 1ms/step - loss: 0.7338 [1m 155/1125 [0m [32m-- [0m [37m------------------ [0m [1m1s [0m 1ms/step - loss: 0.7323 [1m 197/1125 [0m [32m--- [0m [37m----------------- [0m [1m1s [0m 1ms/step - loss: 0.7315 [1m 242/1125 [0m [32m---- [0m [37m---------------- [0m [1m1s [0m 1ms/step - loss: 0.7310 [1m 285/1125 [0m [32m----- [0m [37m--------------- [0m [1m1s [0m 1ms/step - loss: 0.7307 [1m 328/1125 [0m [32m----- [0m [37m--------------- [0m [1m0s [0m 1ms/step - loss: 0.7306 [1m 369/1125 [0m [32m------ [0m [37m-------------- [0m [1m0s [0m 1ms/step - loss: 0.7310 [1m 412/1125 [0m [32m------- [0m [37m------------- [0m [1m0s [0m 1ms/step - loss: 0.7316 [1m 450/1125 [0m [32m-------- [0m [37m------------ [0m [1m0s [0m 1ms/step - loss: 0.7321 [1m 489/1125 [0m [32m-------- [0m [37m------------ [0m [1m0s [0m 1ms/step - loss: 0.7327 [1m 527/1125 [0m [32m--------- [0m [37m----------- [0m [1m0s [0m 1ms/step - loss: 0.7332 [1m 567/1125 [0m [32m---------- [0m [37m---------- [0m [1m0s [0m 1ms/step - loss: 0.7337 [1m 608/1125 [0m [32m---------- [0m [37m---------- [0m [1m0s [0m 1ms/step - loss: 0.7340 [1m 649/1125 [0m [32m----------- [0m [37m--------- [0m [1m0s [0m 1ms/step - loss: 0.7342 [1m 686/1125 [0m [32m------------ [0m [37m-------- [0m [1m0s [0m 1ms/step - loss: 0.7345 [1m 726/1125 [0m [32m------------ [0m [37m-------- [0m [1m0s [0m 1ms/step - loss: 0.7346 [1m 768/1125 [0m [32m------------- [0m [37m------- [0m [1m0s [0m 1ms/step - loss: 0.7348 [1m 809/1125 [0m [32m-------------- [0m [37m------ [0m [1m0s [0m 1ms/step - loss: 0.7349 [1m 852/1125 [0m [32m--------------- [0m [37m----- [0m [1m0s [0m 1ms/step - loss: 0.7350 [1m 893/1125 [0m [32m--------------- [0m [37m----- [0m [1m0s [0m 1ms/step - loss: 0.7351 [1m 934/1125 [0m [32m---------------- [0m [37m---- [0m [1m0s [0m 1ms/step - loss: 0.7352 [1m 979/1125 [0m [32m----------------- [0m [37m--- [0m [1m0s [0m 1ms/step - loss: 0.7352 [1m1028/1125 [0m [32m------------------ [0m [37m-- [0m [1m0s [0m 1ms/step - loss: 0.7352 [1m1077/1125 [0m [32m------------------- [0m [37m- [0m [1m0s [0m 1ms/step - loss: 0.7353 [1m1124/1125 [0m [32m------------------- [0m [37m- [0m [1m0s [0m 1ms/step - loss: 0.7352 [1m1125/1125 [0m [32m-------------------- [0m [37m [0m [1m2s [0m 1ms/step - loss: 0.7352 - val_loss: 0.7476
Epoch 4/15
[1m 1/1125 [0m [37m-------------------- [0m [1m20s [0m 18ms/step - loss: 0.7447 [1m 37/1125 [0m [37m-------------------- [0m [1m1s [0m 1ms/step - loss: 0.7348 [1m 73/1125 [0m [32m- [0m [37m------------------- [0m [1m1s [0m 1ms/step - loss: 0.7401 [1m 109/1125 [0m [32m- [0m [37m------------------- [0m [1m1s [0m 1ms/step - loss: 0.7394 [1m 145/1125 [0m [32m-- [0m [37m------------------ [0m [1m1s [0m 1ms/step - loss: 0.7353 [1m 185/1125 [0m [32m--- [0m [37m----------------- [0m [1m1s [0m 1ms/step - loss: 0.7319 [1m 229/1125 [0m [32m---- [0m [37m---------------- [0m [1m1s [0m 1ms/step - loss: 0.7295 [1m 268/1125 [0m [32m---- [0m [37m---------------- [0m [1m1s [0m 1ms/step - loss: 0.7280 [1m 313/1125 [0m [32m----- [0m [37m--------------- [0m [1m1s [0m 1ms/step - loss: 0.7269 [1m 359/1125 [0m [32m------ [0m [37m-------------- [0m [1m0s [0m 1ms/step - loss: 0.7258 [1m 405/1125 [0m [32m------- [0m [37m------------- [0m [1m0s [0m 1ms/step - loss: 0.7250 [1m 452/1125 [0m [32m-------- [0m [37m------------ [0m [1m0s [0m 1ms/step - loss: 0.7244 [1m 498/1125 [0m [32m-------- [0m [37m------------ [0m [1m0s [0m 1ms/step - loss: 0.7238 [1m 545/1125 [0m [32m--------- [0m [37m----------- [0m [1m0s [0m 1ms/step - loss: 0.7233 [1m 590/1125 [0m [32m---------- [0m [37m---------- [0m [1m0s [0m 1ms/step - loss: 0.7227 [1m 631/1125 [0m [32m----------- [0m [37m--------- [0m [1m0s [0m 1ms/step - loss: 0.7223 [1m 674/1125 [0m [32m----------- [0m [37m--------- [0m [1m0s [0m 1ms/step - loss: 0.7219 [1m 716/1125 [0m [32m------------ [0m [37m-------- [0m [1m0s [0m 1ms/step - loss: 0.7216 [1m 759/1125 [0m [32m------------- [0m [37m------- [0m [1m0s [0m 1ms/step - loss: 0.7212 [1m 803/1125 [0m [32m-------------- [0m [37m------ [0m [1m0s [0m 1ms/step - loss: 0.7208 [1m 846/1125 [0m [32m--------------- [0m [37m----- [0m [1m0s [0m 1ms/step - loss: 0.7205 [1m 891/1125 [0m [32m--------------- [0m [37m----- [0m [1m0s [0m 1ms/step - loss: 0.7203 [1m 931/1125 [0m [32m---------------- [0m [37m---- [0m [1m0s [0m 1ms/step - loss: 0.7200 [1m 975/1125 [0m [32m----------------- [0m [37m--- [0m [1m0s [0m 1ms/step - loss: 0.7198 [1m1024/1125 [0m [32m------------------ [0m [37m-- [0m [1m0s [0m 1ms/step - loss: 0.7195 [1m1072/1125 [0m [32m------------------- [0m [37m- [0m [1m0s [0m 1ms/step - loss: 0.7193 [1m1113/1125 [0m [32m------------------- [0m [37m- [0m [1m0s [0m 1ms/step - loss: 0.7191 [1m1125/1125 [0m [32m-------------------- [0m [37m [0m [1m1s [0m 1ms/step - loss: 0.7190 - val_loss: 0.7368
Epoch 5/15
[1m 1/1125 [0m [37m-------------------- [0m [1m21s [0m 19ms/step - loss: 0.6325 [1m 38/1125 [0m [37m-------------------- [0m [1m1s [0m 1ms/step - loss: 0.6816 [1m 79/1125 [0m [32m- [0m [37m------------------- [0m [1m1s [0m 1ms/step - loss: 0.6843 [1m 118/1125 [0m [32m-- [0m [37m------------------ [0m [1m1s [0m 1ms/step - loss: 0.6859 [1m 158/1125 [0m [32m-- [0m [37m------------------ [0m [1m1s [0m 1ms/step - loss: 0.6863 [1m 196/1125 [0m [32m--- [0m [37m----------------- [0m [1m1s [0m 1ms/step - loss: 0.6864 [1m 235/1125 [0m [32m---- [0m [37m---------------- [0m [1m1s [0m 1ms/step - loss: 0.6865 [1m 273/1125 [0m [32m---- [0m [37m---------------- [0m [1m1s [0m 1ms/step - loss: 0.6865 [1m 312/1125 [0m [32m----- [0m [37m--------------- [0m [1m1s [0m 1ms/step - loss: 0.6866 [1m 353/1125 [0m [32m------ [0m [37m-------------- [0m [1m0s [0m 1ms/step - loss: 0.6867 [1m 395/1125 [0m [32m------- [0m [37m------------- [0m [1m0s [0m 1ms/step - loss: 0.6868 [1m 433/1125 [0m [32m------- [0m [37m------------- [0m [1m0s [0m 1ms/step - loss: 0.6870 [1m 471/1125 [0m [32m-------- [0m [37m------------ [0m [1m0s [0m 1ms/step - loss: 0.6872 [1m 509/1125 [0m [32m--------- [0m [37m----------- [0m [1m0s [0m 1ms/step - loss: 0.6874 [1m 550/1125 [0m [32m--------- [0m [37m----------- [0m [1m0s [0m 1ms/step - loss: 0.6877 [1m 590/1125 [0m [32m---------- [0m [37m---------- [0m [1m0s [0m 1ms/step - loss: 0.6880 [1m 631/1125 [0m [32m----------- [0m [37m--------- [0m [1m0s [0m 1ms/step - loss: 0.6882 [1m 670/1125 [0m [32m----------- [0m [37m--------- [0m [1m0s [0m 1ms/step - loss: 0.6885 [1m 709/1125 [0m [32m------------ [0m [37m-------- [0m [1m0s [0m 1ms/step - loss: 0.6887 [1m 752/1125 [0m [32m------------- [0m [37m------- [0m [1m0s [0m 1ms/step - loss: 0.6889 [1m 796/1125 [0m [32m-------------- [0m [37m------ [0m [1m0s [0m 1ms/step - loss: 0.6891 [1m 838/1125 [0m [32m-------------- [0m [37m------ [0m [1m0s [0m 1ms/step - loss: 0.6894 [1m 883/1125 [0m [32m--------------- [0m [37m----- [0m [1m0s [0m 1ms/step - loss: 0.6897 [1m 927/1125 [0m [32m---------------- [0m [37m---- [0m [1m0s [0m 1ms/step - loss: 0.6899 [1m 970/1125 [0m [32m----------------- [0m [37m--- [0m [1m0s [0m 1ms/step - loss: 0.6902 [1m1014/1125 [0m [32m------------------ [0m [37m-- [0m [1m0s [0m 1ms/step - loss: 0.6904 [1m1060/1125 [0m [32m------------------ [0m [37m-- [0m [1m0s [0m 1ms/step - loss: 0.6907 [1m1106/1125 [0m [32m------------------- [0m [37m- [0m [1m0s [0m 1ms/step - loss: 0.6909 [1m1125/1125 [0m [32m-------------------- [0m [37m [0m [1m2s [0m 1ms/step - loss: 0.6910 - val_loss: 0.7381
Epoch 6/15
[1m 1/1125 [0m [37m-------------------- [0m [1m21s [0m 19ms/step - loss: 0.5890 [1m 37/1125 [0m [37m-------------------- [0m [1m1s [0m 1ms/step - loss: 0.6619 [1m 75/1125 [0m [32m- [0m [37m------------------- [0m [1m1s [0m 1ms/step - loss: 0.6606 [1m 115/1125 [0m [32m-- [0m [37m------------------ [0m [1m1s [0m 1ms/step - loss: 0.6606 [1m 153/1125 [0m [32m-- [0m [37m------------------ [0m [1m1s [0m 1ms/step - loss: 0.6603 [1m 191/1125 [0m [32m--- [0m [37m----------------- [0m [1m1s [0m 1ms/step - loss: 0.6607 [1m 229/1125 [0m [32m---- [0m [37m---------------- [0m [1m1s [0m 1ms/step - loss: 0.6611 [1m 268/1125 [0m [32m---- [0m [37m---------------- [0m [1m1s [0m 1ms/step - loss: 0.6613 [1m 305/1125 [0m [32m----- [0m [37m--------------- [0m [1m1s [0m 1ms/step - loss: 0.6615 [1m 341/1125 [0m [32m------ [0m [37m-------------- [0m [1m1s [0m 1ms/step - loss: 0.6619 [1m 378/1125 [0m [32m------ [0m [37m-------------- [0m [1m1s [0m 1ms/step - loss: 0.6623 [1m 416/1125 [0m [32m------- [0m [37m------------- [0m [1m0s [0m 1ms/step - loss: 0.6628 [1m 456/1125 [0m [32m-------- [0m [37m------------ [0m [1m0s [0m 1ms/step - loss: 0.6633 [1m 494/1125 [0m [32m-------- [0m [37m------------ [0m [1m0s [0m 1ms/step - loss: 0.6638 [1m 532/1125 [0m [32m--------- [0m [37m----------- [0m [1m0s [0m 1ms/step - loss: 0.6642 [1m 570/1125 [0m [32m---------- [0m [37m---------- [0m [1m0s [0m 1ms/step - loss: 0.6647 [1m 609/1125 [0m [32m---------- [0m [37m---------- [0m [1m0s [0m 1ms/step - loss: 0.6651 [1m 647/1125 [0m [32m----------- [0m [37m--------- [0m [1m0s [0m 1ms/step - loss: 0.6655 [1m 684/1125 [0m [32m------------ [0m [37m-------- [0m [1m0s [0m 1ms/step - loss: 0.6660 [1m 721/1125 [0m [32m------------ [0m [37m-------- [0m [1m0s [0m 1ms/step - loss: 0.6665 [1m 759/1125 [0m [32m------------- [0m [37m------- [0m [1m0s [0m 1ms/step - loss: 0.6670 [1m 797/1125 [0m [32m-------------- [0m [37m------ [0m [1m0s [0m 1ms/step - loss: 0.6676 [1m 838/1125 [0m [32m-------------- [0m [37m------ [0m [1m0s [0m 1ms/step - loss: 0.6682 [1m 877/1125 [0m [32m--------------- [0m [37m----- [0m [1m0s [0m 1ms/step - loss: 0.6687 [1m 920/1125 [0m [32m---------------- [0m [37m---- [0m [1m0s [0m 1ms/step - loss: 0.6694 [1m 962/1125 [0m [32m----------------- [0m [37m--- [0m [1m0s [0m 1ms/step - loss: 0.6699 [1m1000/1125 [0m [32m----------------- [0m [37m--- [0m [1m0s [0m 1ms/step - loss: 0.6704 [1m1036/1125 [0m [32m------------------ [0m [37m-- [0m [1m0s [0m 1ms/step - loss: 0.6709 [1m1072/1125 [0m [32m------------------- [0m [37m- [0m [1m0s [0m 1ms/step - loss: 0.6713 [1m1109/1125 [0m [32m------------------- [0m [37m- [0m [1m0s [0m 1ms/step - loss: 0.6717 [1m1125/1125 [0m [32m-------------------- [0m [37m [0m [1m2s [0m 1ms/step - loss: 0.6719 - val_loss: 0.7258
Epoch 7/15
[1m 1/1125 [0m [37m-------------------- [0m [1m21s [0m 19ms/step - loss: 0.7127 [1m 35/1125 [0m [37m-------------------- [0m [1m1s [0m 1ms/step - loss: 0.6413 [1m 71/1125 [0m [32m- [0m [37m------------------- [0m [1m1s [0m 1ms/step - loss: 0.6478 [1m 109/1125 [0m [32m- [0m [37m------------------- [0m [1m1s [0m 1ms/step - loss: 0.6541 [1m 146/1125 [0m [32m-- [0m [37m------------------ [0m [1m1s [0m 1ms/step - loss: 0.6567 [1m 185/1125 [0m [32m--- [0m [37m----------------- [0m [1m1s [0m 1ms/step - loss: 0.6582 [1m 223/1125 [0m [32m--- [0m [37m----------------- [0m [1m1s [0m 1ms/step - loss: 0.6598 [1m 262/1125 [0m [32m---- [0m [37m---------------- [0m [1m1s [0m 1ms/step - loss: 0.6608 [1m 302/1125 [0m [32m----- [0m [37m--------------- [0m [1m1s [0m 1ms/step - loss: 0.6615 [1m 345/1125 [0m [32m------ [0m [37m-------------- [0m [1m1s [0m 1ms/step - loss: 0.6622 [1m 389/1125 [0m [32m------ [0m [37m-------------- [0m [1m0s [0m 1ms/step - loss: 0.6626 [1m 430/1125 [0m [32m------- [0m [37m------------- [0m [1m0s [0m 1ms/step - loss: 0.6630 [1m 472/1125 [0m [32m-------- [0m [37m------------ [0m [1m0s [0m 1ms/step - loss: 0.6634 [1m 517/1125 [0m [32m--------- [0m [37m----------- [0m [1m0s [0m 1ms/step - loss: 0.6639 [1m 561/1125 [0m [32m--------- [0m [37m----------- [0m [1m0s [0m 1ms/step - loss: 0.6643 [1m 599/1125 [0m [32m---------- [0m [37m---------- [0m [1m0s [0m 1ms/step - loss: 0.6647 [1m 637/1125 [0m [32m----------- [0m [37m--------- [0m [1m0s [0m 1ms/step - loss: 0.6650 [1m 677/1125 [0m [32m------------ [0m [37m-------- [0m [1m0s [0m 1ms/step - loss: 0.6654 [1m 717/1125 [0m [32m------------ [0m [37m-------- [0m [1m0s [0m 1ms/step - loss: 0.6657 [1m 755/1125 [0m [32m------------- [0m [37m------- [0m [1m0s [0m 1ms/step - loss: 0.6661 [1m 794/1125 [0m [32m-------------- [0m [37m------ [0m [1m0s [0m 1ms/step - loss: 0.6665 [1m 833/1125 [0m [32m-------------- [0m [37m------ [0m [1m0s [0m 1ms/step - loss: 0.6668 [1m 872/1125 [0m [32m--------------- [0m [37m----- [0m [1m0s [0m 1ms/step - loss: 0.6671 [1m 911/1125 [0m [32m---------------- [0m [37m---- [0m [1m0s [0m 1ms/step - loss: 0.6674 [1m 949/1125 [0m [32m---------------- [0m [37m---- [0m [1m0s [0m 1ms/step - loss: 0.6677 [1m 987/1125 [0m [32m----------------- [0m [37m--- [0m [1m0s [0m 1ms/step - loss: 0.6679 [1m1024/1125 [0m [32m------------------ [0m [37m-- [0m [1m0s [0m 1ms/step - loss: 0.6682 [1m1063/1125 [0m [32m------------------ [0m [37m-- [0m [1m0s [0m 1ms/step - loss: 0.6684 [1m1101/1125 [0m [32m------------------- [0m [37m- [0m [1m0s [0m 1ms/step - loss: 0.6686 [1m1125/1125 [0m [32m-------------------- [0m [37m [0m [1m2s [0m 1ms/step - loss: 0.6687 - val_loss: 0.7297
Epoch 8/15
[1m 1/1125 [0m [37m-------------------- [0m [1m21s [0m 19ms/step - loss: 0.8105 [1m 39/1125 [0m [37m-------------------- [0m [1m1s [0m 1ms/step - loss: 0.6967 [1m 76/1125 [0m [32m- [0m [37m------------------- [0m [1m1s [0m 1ms/step - loss: 0.6795 [1m 113/1125 [0m [32m-- [0m [37m------------------ [0m [1m1s [0m 1ms/step - loss: 0.6704 [1m 150/1125 [0m [32m-- [0m [37m------------------ [0m [1m1s [0m 1ms/step - loss: 0.6645 [1m 190/1125 [0m [32m--- [0m [37m----------------- [0m [1m1s [0m 1ms/step - loss: 0.6622 [1m 228/1125 [0m [32m---- [0m [37m---------------- [0m [1m1s [0m 1ms/step - loss: 0.6613 [1m 267/1125 [0m [32m---- [0m [37m---------------- [0m [1m1s [0m 1ms/step - loss: 0.6607 [1m 306/1125 [0m [32m----- [0m [37m--------------- [0m [1m1s [0m 1ms/step - loss: 0.6604 [1m 346/1125 [0m [32m------ [0m [37m-------------- [0m [1m1s [0m 1ms/step - loss: 0.6602 [1m 387/1125 [0m [32m------ [0m [37m-------------- [0m [1m0s [0m 1ms/step - loss: 0.6600 [1m 426/1125 [0m [32m------- [0m [37m------------- [0m [1m0s [0m 1ms/step - loss: 0.6599 [1m 467/1125 [0m [32m-------- [0m [37m------------ [0m [1m0s [0m 1ms/step - loss: 0.6599 [1m 510/1125 [0m [32m--------- [0m [37m----------- [0m [1m0s [0m 1ms/step - loss: 0.6600 [1m 552/1125 [0m [32m--------- [0m [37m----------- [0m [1m0s [0m 1ms/step - loss: 0.6600 [1m 594/1125 [0m [32m---------- [0m [37m---------- [0m [1m0s [0m 1ms/step - loss: 0.6600 [1m 634/1125 [0m [32m----------- [0m [37m--------- [0m [1m0s [0m 1ms/step - loss: 0.6601 [1m 672/1125 [0m [32m----------- [0m [37m--------- [0m [1m0s [0m 1ms/step - loss: 0.6602 [1m 712/1125 [0m [32m------------ [0m [37m-------- [0m [1m0s [0m 1ms/step - loss: 0.6603 [1m 755/1125 [0m [32m------------- [0m [37m------- [0m [1m0s [0m 1ms/step - loss: 0.6604 [1m 791/1125 [0m [32m-------------- [0m [37m------ [0m [1m0s [0m 1ms/step - loss: 0.6605 [1m 828/1125 [0m [32m-------------- [0m [37m------ [0m [1m0s [0m 1ms/step - loss: 0.6606 [1m 869/1125 [0m [32m--------------- [0m [37m----- [0m [1m0s [0m 1ms/step - loss: 0.6608 [1m 907/1125 [0m [32m---------------- [0m [37m---- [0m [1m0s [0m 1ms/step - loss: 0.6609 [1m 948/1125 [0m [32m---------------- [0m [37m---- [0m [1m0s [0m 1ms/step - loss: 0.6611 [1m 989/1125 [0m [32m----------------- [0m [37m--- [0m [1m0s [0m 1ms/step - loss: 0.6612 [1m1030/1125 [0m [32m------------------ [0m [37m-- [0m [1m0s [0m 1ms/step - loss: 0.6614 [1m1071/1125 [0m [32m------------------- [0m [37m- [0m [1m0s [0m 1ms/step - loss: 0.6615 [1m1114/1125 [0m [32m------------------- [0m [37m- [0m [1m0s [0m 1ms/step - loss: 0.6617 [1m1125/1125 [0m [32m-------------------- [0m [37m [0m [1m2s [0m 1ms/step - loss: 0.6617 - val_loss: 0.7296
Epoch 9/15
[1m 1/1125 [0m [37m-------------------- [0m [1m21s [0m 19ms/step - loss: 0.6438 [1m 41/1125 [0m [37m-------------------- [0m [1m1s [0m 1ms/step - loss: 0.6322 [1m 81/1125 [0m [32m- [0m [37m------------------- [0m [1m1s [0m 1ms/step - loss: 0.6397 [1m 121/1125 [0m [32m-- [0m [37m------------------ [0m [1m1s [0m 1ms/step - loss: 0.6433 [1m 162/1125 [0m [32m-- [0m [37m------------------ [0m [1m1s [0m 1ms/step - loss: 0.6447 [1m 203/1125 [0m [32m--- [0m [37m----------------- [0m [1m1s [0m 1ms/step - loss: 0.6451 [1m 242/1125 [0m [32m---- [0m [37m---------------- [0m [1m1s [0m 1ms/step - loss: 0.6456 [1m 283/1125 [0m [32m----- [0m [37m--------------- [0m [1m1s [0m 1ms/step - loss: 0.6464 [1m 327/1125 [0m [32m----- [0m [37m--------------- [0m [1m0s [0m 1ms/step - loss: 0.6472 [1m 370/1125 [0m [32m------ [0m [37m-------------- [0m [1m0s [0m 1ms/step - loss: 0.6477 [1m 411/1125 [0m [32m------- [0m [37m------------- [0m [1m0s [0m 1ms/step - loss: 0.6483 [1m 450/1125 [0m [32m-------- [0m [37m------------ [0m [1m0s [0m 1ms/step - loss: 0.6487 [1m 489/1125 [0m [32m-------- [0m [37m------------ [0m [1m0s [0m 1ms/step - loss: 0.6491 [1m 531/1125 [0m [32m--------- [0m [37m----------- [0m [1m0s [0m 1ms/step - loss: 0.6494 [1m 572/1125 [0m [32m---------- [0m [37m---------- [0m [1m0s [0m 1ms/step - loss: 0.6496 [1m 611/1125 [0m [32m---------- [0m [37m---------- [0m [1m0s [0m 1ms/step - loss: 0.6498 [1m 650/1125 [0m [32m----------- [0m [37m--------- [0m [1m0s [0m 1ms/step - loss: 0.6500 [1m 689/1125 [0m [32m------------ [0m [37m-------- [0m [1m0s [0m 1ms/step - loss: 0.6502 [1m 727/1125 [0m [32m------------ [0m [37m-------- [0m [1m0s [0m 1ms/step - loss: 0.6504 [1m 767/1125 [0m [32m------------- [0m [37m------- [0m [1m0s [0m 1ms/step - loss: 0.6507 [1m 805/1125 [0m [32m-------------- [0m [37m------ [0m [1m0s [0m 1ms/step - loss: 0.6509 [1m 843/1125 [0m [32m-------------- [0m [37m------ [0m [1m0s [0m 1ms/step - loss: 0.6511 [1m 883/1125 [0m [32m--------------- [0m [37m----- [0m [1m0s [0m 1ms/step - loss: 0.6514 [1m 922/1125 [0m [32m---------------- [0m [37m---- [0m [1m0s [0m 1ms/step - loss: 0.6516 [1m 966/1125 [0m [32m----------------- [0m [37m--- [0m [1m0s [0m 1ms/step - loss: 0.6519 [1m1004/1125 [0m [32m----------------- [0m [37m--- [0m [1m0s [0m 1ms/step - loss: 0.6521 [1m1044/1125 [0m [32m------------------ [0m [37m-- [0m [1m0s [0m 1ms/step - loss: 0.6524 [1m1085/1125 [0m [32m------------------- [0m [37m- [0m [1m0s [0m 1ms/step - loss: 0.6526 [1m1125/1125 [0m [32m-------------------- [0m [37m [0m [1m2s [0m 1ms/step - loss: 0.6528 - val_loss: 0.7266
Epoch 10/15
[1m 1/1125 [0m [37m-------------------- [0m [1m20s [0m 18ms/step - loss: 0.6382 [1m 37/1125 [0m [37m-------------------- [0m [1m1s [0m 1ms/step - loss: 0.6338 [1m 73/1125 [0m [32m- [0m [37m------------------- [0m [1m1s [0m 1ms/step - loss: 0.6296 [1m 109/1125 [0m [32m- [0m [37m------------------- [0m [1m1s [0m 1ms/step - loss: 0.6265 [1m 148/1125 [0m [32m-- [0m [37m------------------ [0m [1m1s [0m 1ms/step - loss: 0.6269 [1m 187/1125 [0m [32m--- [0m [37m----------------- [0m [1m1s [0m 1ms/step - loss: 0.6284 [1m 225/1125 [0m [32m---- [0m [37m---------------- [0m [1m1s [0m 1ms/step - loss: 0.6303 [1m 263/1125 [0m [32m---- [0m [37m---------------- [0m [1m1s [0m 1ms/step - loss: 0.6321 [1m 302/1125 [0m [32m----- [0m [37m--------------- [0m [1m1s [0m 1ms/step - loss: 0.6333 [1m 339/1125 [0m [32m------ [0m [37m-------------- [0m [1m1s [0m 1ms/step - loss: 0.6342 [1m 377/1125 [0m [32m------ [0m [37m-------------- [0m [1m1s [0m 1ms/step - loss: 0.6351 [1m 414/1125 [0m [32m------- [0m [37m------------- [0m [1m0s [0m 1ms/step - loss: 0.6358 [1m 453/1125 [0m [32m-------- [0m [37m------------ [0m [1m0s [0m 1ms/step - loss: 0.6365 [1m 495/1125 [0m [32m-------- [0m [37m------------ [0m [1m0s [0m 1ms/step - loss: 0.6371 [1m 535/1125 [0m [32m--------- [0m [37m----------- [0m [1m0s [0m 1ms/step - loss: 0.6376 [1m 576/1125 [0m [32m---------- [0m [37m---------- [0m [1m0s [0m 1ms/step - loss: 0.6380 [1m 617/1125 [0m [32m---------- [0m [37m---------- [0m [1m0s [0m 1ms/step - loss: 0.6385 [1m 660/1125 [0m [32m----------- [0m [37m--------- [0m [1m0s [0m 1ms/step - loss: 0.6389 [1m 704/1125 [0m [32m------------ [0m [37m-------- [0m [1m0s [0m 1ms/step - loss: 0.6393 [1m 746/1125 [0m [32m------------- [0m [37m------- [0m [1m0s [0m 1ms/step - loss: 0.6397 [1m 790/1125 [0m [32m-------------- [0m [37m------ [0m [1m0s [0m 1ms/step - loss: 0.6400 [1m 831/1125 [0m [32m-------------- [0m [37m------ [0m [1m0s [0m 1ms/step - loss: 0.6404 [1m 873/1125 [0m [32m--------------- [0m [37m----- [0m [1m0s [0m 1ms/step - loss: 0.6408 [1m 915/1125 [0m [32m---------------- [0m [37m---- [0m [1m0s [0m 1ms/step - loss: 0.6411 [1m 957/1125 [0m [32m----------------- [0m [37m--- [0m [1m0s [0m 1ms/step - loss: 0.6415 [1m1001/1125 [0m [32m----------------- [0m [37m--- [0m [1m0s [0m 1ms/step - loss: 0.6419 [1m1047/1125 [0m [32m------------------ [0m [37m-- [0m [1m0s [0m 1ms/step - loss: 0.6423 [1m1092/1125 [0m [32m------------------- [0m [37m- [0m [1m0s [0m 1ms/step - loss: 0.6427 [1m1125/1125 [0m [32m-------------------- [0m [37m [0m [1m2s [0m 1ms/step - loss: 0.6430 - val_loss: 0.7293
Epoch 11/15
[1m 1/1125 [0m [37m-------------------- [0m [1m21s [0m 19ms/step - loss: 0.6002 [1m 42/1125 [0m [37m-------------------- [0m [1m1s [0m 1ms/step - loss: 0.6190 [1m 91/1125 [0m [32m- [0m [37m------------------- [0m [1m1s [0m 1ms/step - loss: 0.6249 [1m 144/1125 [0m [32m-- [0m [37m------------------ [0m [1m1s [0m 1ms/step - loss: 0.6274 [1m 197/1125 [0m [32m--- [0m [37m----------------- [0m [1m0s [0m 1ms/step - loss: 0.6282 [1m 250/1125 [0m [32m---- [0m [37m---------------- [0m [1m0s [0m 1ms/step - loss: 0.6295 [1m 304/1125 [0m [32m----- [0m [37m--------------- [0m [1m0s [0m 1000us/step - loss: 0.6310 [1m 356/1125 [0m [32m------ [0m [37m-------------- [0m [1m0s [0m 995us/step - loss: 0.6324 [1m 411/1125 [0m [32m------- [0m [37m------------- [0m [1m0s [0m 986us/step - loss: 0.6337 [1m 456/1125 [0m [32m-------- [0m [37m------------ [0m [1m0s [0m 998us/step - loss: 0.6346 [1m 496/1125 [0m [32m-------- [0m [37m------------ [0m [1m0s [0m 1ms/step - loss: 0.6356 [1m 540/1125 [0m [32m--------- [0m [37m----------- [0m [1m0s [0m 1ms/step - loss: 0.6365 [1m 584/1125 [0m [32m---------- [0m [37m---------- [0m [1m0s [0m 1ms/step - loss: 0.6373 [1m 627/1125 [0m [32m----------- [0m [37m--------- [0m [1m0s [0m 1ms/step - loss: 0.6379 [1m 669/1125 [0m [32m----------- [0m [37m--------- [0m [1m0s [0m 1ms/step - loss: 0.6384 [1m 712/1125 [0m [32m------------ [0m [37m-------- [0m [1m0s [0m 1ms/step - loss: 0.6389 [1m 753/1125 [0m [32m------------- [0m [37m------- [0m [1m0s [0m 1ms/step - loss: 0.6392 [1m 792/1125 [0m [32m-------------- [0m [37m------ [0m [1m0s [0m 1ms/step - loss: 0.6395 [1m 832/1125 [0m [32m-------------- [0m [37m------ [0m [1m0s [0m 1ms/step - loss: 0.6397 [1m 871/1125 [0m [32m--------------- [0m [37m----- [0m [1m0s [0m 1ms/step - loss: 0.6400 [1m 910/1125 [0m [32m---------------- [0m [37m---- [0m [1m0s [0m 1ms/step - loss: 0.6402 [1m 949/1125 [0m [32m---------------- [0m [37m---- [0m [1m0s [0m 1ms/step - loss: 0.6404 [1m 986/1125 [0m [32m----------------- [0m [37m--- [0m [1m0s [0m 1ms/step - loss: 0.6406 [1m1024/1125 [0m [32m------------------ [0m [37m-- [0m [1m0s [0m 1ms/step - loss: 0.6407 [1m1064/1125 [0m [32m------------------ [0m [37m-- [0m [1m0s [0m 1ms/step - loss: 0.6409 [1m1104/1125 [0m [32m------------------- [0m [37m- [0m [1m0s [0m 1ms/step - loss: 0.6411 [1m1125/1125 [0m [32m-------------------- [0m [37m [0m [1m1s [0m 1ms/step - loss: 0.6413 - val_loss: 0.7222
Epoch 12/15
[1m 1/1125 [0m [37m-------------------- [0m [1m20s [0m 18ms/step - loss: 0.5863 [1m 35/1125 [0m [37m-------------------- [0m [1m1s [0m 1ms/step - loss: 0.6016 [1m 72/1125 [0m [32m- [0m [37m------------------- [0m [1m1s [0m 1ms/step - loss: 0.6089 [1m 108/1125 [0m [32m- [0m [37m------------------- [0m [1m1s [0m 1ms/step - loss: 0.6143 [1m 147/1125 [0m [32m-- [0m [37m------------------ [0m [1m1s [0m 1ms/step - loss: 0.6186 [1m 186/1125 [0m [32m--- [0m [37m----------------- [0m [1m1s [0m 1ms/step - loss: 0.6209 [1m 226/1125 [0m [32m---- [0m [37m---------------- [0m [1m1s [0m 1ms/step - loss: 0.6233 [1m 264/1125 [0m [32m---- [0m [37m---------------- [0m [1m1s [0m 1ms/step - loss: 0.6250 [1m 304/1125 [0m [32m----- [0m [37m--------------- [0m [1m1s [0m 1ms/step - loss: 0.6266 [1m 343/1125 [0m [32m------ [0m [37m-------------- [0m [1m1s [0m 1ms/step - loss: 0.6277 [1m 382/1125 [0m [32m------ [0m [37m-------------- [0m [1m0s [0m 1ms/step - loss: 0.6283 [1m 419/1125 [0m [32m------- [0m [37m------------- [0m [1m0s [0m 1ms/step - loss: 0.6288 [1m 457/1125 [0m [32m-------- [0m [37m------------ [0m [1m0s [0m 1ms/step - loss: 0.6291 [1m 494/1125 [0m [32m-------- [0m [37m------------ [0m [1m0s [0m 1ms/step - loss: 0.6294 [1m 536/1125 [0m [32m--------- [0m [37m----------- [0m [1m0s [0m 1ms/step - loss: 0.6299 [1m 577/1125 [0m [32m---------- [0m [37m---------- [0m [1m0s [0m 1ms/step - loss: 0.6303 [1m 618/1125 [0m [32m---------- [0m [37m---------- [0m [1m0s [0m 1ms/step - loss: 0.6306 [1m 657/1125 [0m [32m----------- [0m [37m--------- [0m [1m0s [0m 1ms/step - loss: 0.6310 [1m 694/1125 [0m [32m------------ [0m [37m-------- [0m [1m0s [0m 1ms/step - loss: 0.6314 [1m 731/1125 [0m [32m------------ [0m [37m-------- [0m [1m0s [0m 1ms/step - loss: 0.6317 [1m 769/1125 [0m [32m------------- [0m [37m------- [0m [1m0s [0m 1ms/step - loss: 0.6320 [1m 811/1125 [0m [32m-------------- [0m [37m------ [0m [1m0s [0m 1ms/step - loss: 0.6323 [1m 855/1125 [0m [32m--------------- [0m [37m----- [0m [1m0s [0m 1ms/step - loss: 0.6327 [1m 893/1125 [0m [32m--------------- [0m [37m----- [0m [1m0s [0m 1ms/step - loss: 0.6329 [1m 929/1125 [0m [32m---------------- [0m [37m---- [0m [1m0s [0m 1ms/step - loss: 0.6332 [1m 967/1125 [0m [32m----------------- [0m [37m--- [0m [1m0s [0m 1ms/step - loss: 0.6335 [1m1006/1125 [0m [32m----------------- [0m [37m--- [0m [1m0s [0m 1ms/step - loss: 0.6338 [1m1045/1125 [0m [32m------------------ [0m [37m-- [0m [1m0s [0m 1ms/step - loss: 0.6341 [1m1083/1125 [0m [32m------------------- [0m [37m- [0m [1m0s [0m 1ms/step - loss: 0.6343 [1m1124/1125 [0m [32m------------------- [0m [37m- [0m [1m0s [0m 1ms/step - loss: 0.6346 [1m1125/1125 [0m [32m-------------------- [0m [37m [0m [1m2s [0m 1ms/step - loss: 0.6346 - val_loss: 0.7320
Epoch 13/15
[1m 1/1125 [0m [37m-------------------- [0m [1m20s [0m 18ms/step - loss: 0.7768 [1m 42/1125 [0m [37m-------------------- [0m [1m1s [0m 1ms/step - loss: 0.6361 [1m 80/1125 [0m [32m- [0m [37m------------------- [0m [1m1s [0m 1ms/step - loss: 0.6281 [1m 118/1125 [0m [32m-- [0m [37m------------------ [0m [1m1s [0m 1ms/step - loss: 0.6259 [1m 157/1125 [0m [32m-- [0m [37m------------------ [0m [1m1s [0m 1ms/step - loss: 0.6257 [1m 194/1125 [0m [32m--- [0m [37m----------------- [0m [1m1s [0m 1ms/step - loss: 0.6255 [1m 233/1125 [0m [32m---- [0m [37m---------------- [0m [1m1s [0m 1ms/step - loss: 0.6250 [1m 273/1125 [0m [32m---- [0m [37m---------------- [0m [1m1s [0m 1ms/step - loss: 0.6250 [1m 314/1125 [0m [32m----- [0m [37m--------------- [0m [1m1s [0m 1ms/step - loss: 0.6252 [1m 356/1125 [0m [32m------ [0m [37m-------------- [0m [1m0s [0m 1ms/step - loss: 0.6255 [1m 397/1125 [0m [32m------- [0m [37m------------- [0m [1m0s [0m 1ms/step - loss: 0.6259 [1m 438/1125 [0m [32m------- [0m [37m------------- [0m [1m0s [0m 1ms/step - loss: 0.6263 [1m 477/1125 [0m [32m-------- [0m [37m------------ [0m [1m0s [0m 1ms/step - loss: 0.6266 [1m 518/1125 [0m [32m--------- [0m [37m----------- [0m [1m0s [0m 1ms/step - loss: 0.6269 [1m 559/1125 [0m [32m--------- [0m [37m----------- [0m [1m0s [0m 1ms/step - loss: 0.6271 [1m 599/1125 [0m [32m---------- [0m [37m---------- [0m [1m0s [0m 1ms/step - loss: 0.6273 [1m 637/1125 [0m [32m----------- [0m [37m--------- [0m [1m0s [0m 1ms/step - loss: 0.6275 [1m 675/1125 [0m [32m------------ [0m [37m-------- [0m [1m0s [0m 1ms/step - loss: 0.6277 [1m 714/1125 [0m [32m------------ [0m [37m-------- [0m [1m0s [0m 1ms/step - loss: 0.6279 [1m 754/1125 [0m [32m------------- [0m [37m------- [0m [1m0s [0m 1ms/step - loss: 0.6280 [1m 793/1125 [0m [32m-------------- [0m [37m------ [0m [1m0s [0m 1ms/step - loss: 0.6281 [1m 832/1125 [0m [32m-------------- [0m [37m------ [0m [1m0s [0m 1ms/step - loss: 0.6283 [1m 870/1125 [0m [32m--------------- [0m [37m----- [0m [1m0s [0m 1ms/step - loss: 0.6285 [1m 908/1125 [0m [32m---------------- [0m [37m---- [0m [1m0s [0m 1ms/step - loss: 0.6287 [1m 946/1125 [0m [32m---------------- [0m [37m---- [0m [1m0s [0m 1ms/step - loss: 0.6290 [1m 982/1125 [0m [32m----------------- [0m [37m--- [0m [1m0s [0m 1ms/step - loss: 0.6292 [1m1020/1125 [0m [32m------------------ [0m [37m-- [0m [1m0s [0m 1ms/step - loss: 0.6295 [1m1059/1125 [0m [32m------------------ [0m [37m-- [0m [1m0s [0m 1ms/step - loss: 0.6298 [1m1096/1125 [0m [32m------------------- [0m [37m- [0m [1m0s [0m 1ms/step - loss: 0.6300 [1m1125/1125 [0m [32m-------------------- [0m [37m [0m [1m2s [0m 1ms/step - loss: 0.6303 - val_loss: 0.7281
Epoch 14/15
[1m 1/1125 [0m [37m-------------------- [0m [1m21s [0m 19ms/step - loss: 0.7193 [1m 42/1125 [0m [37m-------------------- [0m [1m1s [0m 1ms/step - loss: 0.6342 [1m 85/1125 [0m [32m- [0m [37m------------------- [0m [1m1s [0m 1ms/step - loss: 0.6372 [1m 126/1125 [0m [32m-- [0m [37m------------------ [0m [1m1s [0m 1ms/step - loss: 0.6361 [1m 168/1125 [0m [32m-- [0m [37m------------------ [0m [1m1s [0m 1ms/step - loss: 0.6352 [1m 208/1125 [0m [32m--- [0m [37m----------------- [0m [1m1s [0m 1ms/step - loss: 0.6347 [1m 251/1125 [0m [32m---- [0m [37m---------------- [0m [1m1s [0m 1ms/step - loss: 0.6339 [1m 294/1125 [0m [32m----- [0m [37m--------------- [0m [1m1s [0m 1ms/step - loss: 0.6332 [1m 336/1125 [0m [32m----- [0m [37m--------------- [0m [1m0s [0m 1ms/step - loss: 0.6325 [1m 377/1125 [0m [32m------ [0m [37m-------------- [0m [1m0s [0m 1ms/step - loss: 0.6320 [1m 419/1125 [0m [32m------- [0m [37m------------- [0m [1m0s [0m 1ms/step - loss: 0.6315 [1m 461/1125 [0m [32m-------- [0m [37m------------ [0m [1m0s [0m 1ms/step - loss: 0.6312 [1m 502/1125 [0m [32m-------- [0m [37m------------ [0m [1m0s [0m 1ms/step - loss: 0.6309 [1m 543/1125 [0m [32m--------- [0m [37m----------- [0m [1m0s [0m 1ms/step - loss: 0.6308 [1m 586/1125 [0m [32m---------- [0m [37m---------- [0m [1m0s [0m 1ms/step - loss: 0.6306 [1m 623/1125 [0m [32m----------- [0m [37m--------- [0m [1m0s [0m 1ms/step - loss: 0.6306 [1m 662/1125 [0m [32m----------- [0m [37m--------- [0m [1m0s [0m 1ms/step - loss: 0.6305 [1m 703/1125 [0m [32m------------ [0m [37m-------- [0m [1m0s [0m 1ms/step - loss: 0.6305 [1m 744/1125 [0m [32m------------- [0m [37m------- [0m [1m0s [0m 1ms/step - loss: 0.6306 [1m 789/1125 [0m [32m-------------- [0m [37m------ [0m [1m0s [0m 1ms/step - loss: 0.6307 [1m 832/1125 [0m [32m-------------- [0m [37m------ [0m [1m0s [0m 1ms/step - loss: 0.6307 [1m 877/1125 [0m [32m--------------- [0m [37m----- [0m [1m0s [0m 1ms/step - loss: 0.6308 [1m 922/1125 [0m [32m---------------- [0m [37m---- [0m [1m0s [0m 1ms/step - loss: 0.6308 [1m 967/1125 [0m [32m----------------- [0m [37m--- [0m [1m0s [0m 1ms/step - loss: 0.6309 [1m1013/1125 [0m [32m------------------ [0m [37m-- [0m [1m0s [0m 1ms/step - loss: 0.6310 [1m1057/1125 [0m [32m------------------ [0m [37m-- [0m [1m0s [0m 1ms/step - loss: 0.6311 [1m1103/1125 [0m [32m------------------- [0m [37m- [0m [1m0s [0m 1ms/step - loss: 0.6312 [1m1125/1125 [0m [32m-------------------- [0m [37m [0m [1m2s [0m 1ms/step - loss: 0.6313 - val_loss: 0.7276
Epoch 15/15
[1m 1/1125 [0m [37m-------------------- [0m [1m20s [0m 18ms/step - loss: 0.7280 [1m 43/1125 [0m [37m-------------------- [0m [1m1s [0m 1ms/step - loss: 0.6320 [1m 85/1125 [0m [32m- [0m [37m------------------- [0m [1m1s [0m 1ms/step - loss: 0.6322 [1m 127/1125 [0m [32m-- [0m [37m------------------ [0m [1m1s [0m 1ms/step - loss: 0.6288 [1m 164/1125 [0m [32m-- [0m [37m------------------ [0m [1m1s [0m 1ms/step - loss: 0.6276 [1m 201/1125 [0m [32m--- [0m [37m----------------- [0m [1m1s [0m 1ms/step - loss: 0.6269 [1m 238/1125 [0m [32m---- [0m [37m---------------- [0m [1m1s [0m 1ms/step - loss: 0.6264 [1m 275/1125 [0m [32m---- [0m [37m---------------- [0m [1m1s [0m 1ms/step - loss: 0.6258 [1m 311/1125 [0m [32m----- [0m [37m--------------- [0m [1m1s [0m 1ms/step - loss: 0.6254 [1m 350/1125 [0m [32m------ [0m [37m-------------- [0m [1m1s [0m 1ms/step - loss: 0.6251 [1m 389/1125 [0m [32m------ [0m [37m-------------- [0m [1m0s [0m 1ms/step - loss: 0.6250 [1m 429/1125 [0m [32m------- [0m [37m------------- [0m [1m0s [0m 1ms/step - loss: 0.6250 [1m 470/1125 [0m [32m-------- [0m [37m------------ [0m [1m0s [0m 1ms/step - loss: 0.6252 [1m 510/1125 [0m [32m--------- [0m [37m----------- [0m [1m0s [0m 1ms/step - loss: 0.6253 [1m 549/1125 [0m [32m--------- [0m [37m----------- [0m [1m0s [0m 1ms/step - loss: 0.6255 [1m 590/1125 [0m [32m---------- [0m [37m---------- [0m [1m0s [0m 1ms/step - loss: 0.6256 [1m 631/1125 [0m [32m----------- [0m [37m--------- [0m [1m0s [0m 1ms/step - loss: 0.6256 [1m 673/1125 [0m [32m----------- [0m [37m--------- [0m [1m0s [0m 1ms/step - loss: 0.6257 [1m 712/1125 [0m [32m------------ [0m [37m-------- [0m [1m0s [0m 1ms/step - loss: 0.6258 [1m 751/1125 [0m [32m------------- [0m [37m------- [0m [1m0s [0m 1ms/step - loss: 0.6259 [1m 796/1125 [0m [32m-------------- [0m [37m------ [0m [1m0s [0m 1ms/step - loss: 0.6259 [1m 837/1125 [0m [32m-------------- [0m [37m------ [0m [1m0s [0m 1ms/step - loss: 0.6260 [1m 877/1125 [0m [32m--------------- [0m [37m----- [0m [1m0s [0m 1ms/step - loss: 0.6260 [1m 918/1125 [0m [32m---------------- [0m [37m---- [0m [1m0s [0m 1ms/step - loss: 0.6260 [1m 962/1125 [0m [32m----------------- [0m [37m--- [0m [1m0s [0m 1ms/step - loss: 0.6261 [1m1002/1125 [0m [32m----------------- [0m [37m--- [0m [1m0s [0m 1ms/step - loss: 0.6262 [1m1043/1125 [0m [32m------------------ [0m [37m-- [0m [1m0s [0m 1ms/step - loss: 0.6263 [1m1085/1125 [0m [32m------------------- [0m [37m- [0m [1m0s [0m 1ms/step - loss: 0.6264 [1m1125/1125 [0m [32m-------------------- [0m [37m [0m [1m2s [0m 1ms/step - loss: 0.6265 - val_loss: 0.7282
[1m 1/625 [0m [37m-------------------- [0m [1m8s [0m 14ms/step [1m 64/625 [0m [32m-- [0m [37m------------------ [0m [1m0s [0m 799us/step [1m127/625 [0m [32m---- [0m [37m---------------- [0m [1m0s [0m 799us/step [1m197/625 [0m [32m------ [0m [37m-------------- [0m [1m0s [0m 771us/step [1m266/625 [0m [32m-------- [0m [37m------------ [0m [1m0s [0m 760us/step [1m336/625 [0m [32m---------- [0m [37m---------- [0m [1m0s [0m 753us/step [1m411/625 [0m [32m------------- [0m [37m------- [0m [1m0s [0m 737us/step [1m487/625 [0m [32m--------------- [0m [37m----- [0m [1m0s [0m 725us/step [1m560/625 [0m [32m----------------- [0m [37m--- [0m [1m0s [0m 720us/step [1m625/625 [0m [32m-------------------- [0m [37m [0m [1m0s [0m 736us/step
Final test Loss: 0.960
Final test Loss: 0.712Test verisi üzerinde hata karelerinin ortalaması (mean squared error) ve hata mutlak değer ortalaması (mean absolute error) görülüyor. Notların 1 ile 5 arasında olduğunu düşünürsek sonuçlar fena değil diyebiliriz.
Tavsiye Üretmek
Eğitilmiş model sayesinde artık kullanıcının hiç seyretmediği filmler için beğeni notu üretmek mümkün, herhangi bir kullanıcı / film kimliğini DYSA’ya soldan veriyoruz, sağdan tahmin notu çıkıyor, DYSA’nın her kullanıcı ve her film için elinde gömme temsili vektörleri var. O zaman tavsiye üretme amacıyla herhangi bir kullanıcı için tüm filmlerin tahmini notlarını hesaplarsak, ve sonuç beğeni notları içinde en üst 10 filmi seçersek bu filmler bu kullanıcının görüp beğenebileceği filmler olacaktır. Kullanıcı 100 için bu tahminleri üretelim,
def recommend(user_id, top_n=10):
all_possible_item_ids = np.arange(1, max_item_id + 1)
seen_movies = set(all_ratings[all_ratings["user_id"] == user_id]["item_id"])
item_ids_to_predict = np.array(list(filter(lambda x: x not in seen_movies, all_possible_item_ids)))
if len(item_ids_to_predict) == 0:
print(f"User {user_id} has seen all available movies or no items to recommend.")
return []
user_array = np.full_like(item_ids_to_predict, user_id).reshape(-1, 1) # Reshape here too
items_meta = items.loc[item_ids_to_predict, "scaled_date"].values.reshape(-1, 1)
rating_preds = model.predict([user_array, item_ids_to_predict.reshape(-1, 1), items_meta]) # Reshape item_ids_to_predict
sorted_indices = np.argsort(rating_preds[:, 0])[::-1]
top_n_item_ids = item_ids_to_predict[sorted_indices[:top_n]]
recommendations = []
for item_id in top_n_item_ids:
movie_name = items.loc[item_id, "name"]
original_idx_in_preds = np.where(item_ids_to_predict == item_id)[0][0]
predicted_rating = rating_preds[original_idx_in_preds][0]
recommendations.append((movie_name, predicted_rating))
return recommendations
for x in recommend(100):
print(x) [1m 1/51 [0m [37m-------------------- [0m [...
('Golden Earrings (1947)', np.float32(5.3875165))
('Delta of Venus (1994)', np.float32(5.0874066))
('Santa with Muscles (1996)', np.float32(4.9496017))
('Hugo Pool (1997)', np.float32(4.886615))
('Little City (1998)', np.float32(4.840723))
('Pather Panchali (1955)', np.float32(4.7911043))
('Mondo (1996)', np.float32(4.6946745))
('Great Day in Harlem, A (1994)', np.float32(4.651801))
('Anna (1996)', np.float32(4.6199737))
('Mina Tannenbaum (1994)', np.float32(4.6016974))Bu arkadaş üstteki filmleri sevebilirmiş.
Yakınlık
Gömme tabakasında her kullanıcı, film daha az boyutta, sayısal değerlerle temsil edilir dedik, mesela Toy Story filmi için
weights = model.get_weights()
user_embeddings = weights[0]
item_embeddings = weights[1]
print("First item name from metadata:", items["name"][1])
print("Embedding vector for the first item:")
print(item_embeddings[1])
print("shape:", item_embeddings[1].shape)First item name from metadata: Toy Story (1995)
Embedding vector for the first item:
[ 0.19047019 0.07151242 -0.14766356 0.15161663 -0.02006965 -0.11304317
-0.01237285 -0.03916255 0.13233434 -0.1622715 0.04152166 -0.21307035
-0.18624651 -0.08918868 0.26397684 -0.08177959 -0.24913797 0.17463812
-0.12850331 0.03812746 0.14893413 -0.17002341 -0.13132991 0.17275381
0.19235374 0.10746008 0.14028865 -0.08930523 -0.23663597 -0.07991344
-0.14265846 -0.15461077]
shape: (32,)Bu sayıların yakınlık bağlamında anlamı olduğundan bahsettik. O zaman bir filmin diğerlerine olan yakınlığını Öklitsel uzaklık ile hesaplayabiliriz, mesela Groundhog Day filmine yakın filmler hangileridir?
def euclidean_distances(x):
return np.linalg.norm(item_embeddings - x, axis=1)
def most_similar(idx, top_n=10, mode='euclidean'):
sorted_indexes=0
dists = euclidean_distances(item_embeddings[idx])
sorted_indexes = np.argsort(dists)
idxs = sorted_indexes[0:top_n]
return list(zip(items["name"][idxs], dists[idxs]))
euc_dists = euclidean_distances(item_embeddings[1])
print(euc_dists.shape)
print(euc_dists[1:5])
print()
print("Groundhog Day filmine yakin filmler:")
for title, dist in most_similar(202, mode="euclidean"):
print(title, dist)(1683,)
[0. 0.9378361 1.0300627 0.6592185]
Groundhog Day filmine yakin filmler:
Groundhog Day (1993) 0.0
Kissed (1996) 0.34770346
Age of Innocence, The (1993) 0.36420903
Walk in the Clouds, A (1995) 0.364914
Don Juan DeMarco (1995) 0.37850082
While You Were Sleeping (1995) 0.3789333
Panther (1995) 0.38063824
Species (1995) 0.38847828
Mask, The (1994) 0.392206
Queen Margot (Reine Margot, La) (1994) 0.39537832Pek anlamsız değil. Tabii girdi verisi ne kadar fazla olursa sonuçlar o kadar iyi olur, kullandığımız veri Movielens 100K verisi, 100,000 tane not var. MovieLens 1M (1 milyon not) ile daha iyi sonuçlar alınabilir (bu veri İstatistik, Pivotlama yazısının dizininde). Okur için ödev olsun.
Kaynaklar
[1] Grisel, Deep Learning, course @ Télécom ParisTech, https://github.com/petermartigny/Deep-Learning
[2] https://keras.io/examples/structured_data/collaborative_filtering_movielens/
[3] https://ogrisel.github.io/decks/2017_dotai_neural_recsys/#1