#NCF Hyperparams
r_emb_dim = 32
r_lr = 0.0005
r_epochs = 10
r_l2 = 0.0000
r_dropout = 0.0
r_batch_size = 40960
r_dense_struct = [16, 4]
#Model instantiation
recommender = Recommender(n_users, n_movies, r_emb_dim, r_dense_struct,
r_dropout, r_l2)
recommender.compile(tf.keras.optimizers.Adam(r_lr),
tf.keras.losses.MeanSquaredError(),
metrics=[tf.keras.metrics.RootMeanSquaredError()])
recommender.train_on_batch(train_ds.batch(r_batch_size))
print(recommender.summary())
#Model fit
recommender.fit(train_ds.batch(r_batch_size),
epochs=r_epochs,
validation_data=eval_ds)
recommender.evaluate(test_ds)
#Lets predict someones
pred = recommender.predict(test_ds)
target = [target for sample, target in test_ds]
sns.distplot(pred, bins=10)
sns.distplot(target, bins=10)
from EDA import EDA
train_ds, eval_ds, test_ds, n_users, n_movies = EDA()
#Matrix Factorizer Hyperparams
f_emb_dim = 16
f_lr = 0.0015
f_epochs = 10
f_batch_size = 40960
#Model instantiation
factorizer = MatrixFactorizer(n_users, n_movies, f_emb_dim)
factorizer.compile(tf.keras.optimizers.Adam(f_lr),
tf.keras.losses.MeanSquaredError(),
metrics=[tf.keras.metrics.RootMeanSquaredError()])
factorizer.train_on_batch(train_ds.batch(f_batch_size))
print(factorizer.summary())
#Model fitting
factorizer.fit(train_ds.batch(f_batch_size),
epochs=f_epochs,
validation_data=eval_ds)
#Test Performance of Factorizer
factorizer.evaluate(test_ds)
#Lets predict someones
pred = factorizer.predict(test_ds)
target = [target for sample, target in test_ds]
sns.distplot(pred, bins=10)