recommenders/models/vae/multinomial_vae.py [348:457]: - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - def _take_sample(self, args): """Sample epsilon ∼ N (0,I) and compute z via reparametrization trick.""" """Calculate latent vector using the reparametrization trick. The idea is that sampling from N (_mean, _var) is s the same as sampling from _mean+ epsilon * _var where epsilon ∼ N(0,I).""" # _mean and _log_var calculated in encoder _mean, _log_var = args # epsilon epsilon = K.random_normal( shape=(K.shape(_mean)[0], self.latent_dim), mean=0.0, stddev=1.0, seed=self.seed, ) return _mean + K.exp(_log_var / 2) * epsilon def nn_batch_generator(self, x_train): """Used for splitting dataset in batches. Args: x_train (numpy.ndarray): The click matrix for the train set, with float values. """ # Shuffle the batch np.random.seed(self.seed) shuffle_index = np.arange(np.shape(x_train)[0]) np.random.shuffle(shuffle_index) x = x_train[shuffle_index, :] y = x_train[shuffle_index, :] # Iterate until making a full epoch counter = 0 while 1: index_batch = shuffle_index[ self.batch_size * counter : self.batch_size * (counter + 1) ] # Decompress batch x_batch = x[index_batch, :] y_batch = y[index_batch, :] counter += 1 yield (np.array(x_batch), np.array(y_batch)) # Stopping rule if counter >= self.number_of_batches: counter = 0 def fit(self, x_train, x_valid, x_val_tr, x_val_te, mapper): """Fit model with the train sets and validate on the validation set. Args: x_train (numpy.ndarray): the click matrix for the train set. x_valid (numpy.ndarray): the click matrix for the validation set. x_val_tr (numpy.ndarray): the click matrix for the validation set training part. x_val_te (numpy.ndarray): the click matrix for the validation set testing part. mapper (object): the mapper for converting click matrix to dataframe. It can be AffinityMatrix. """ # initialise LossHistory used for saving loss of validation and train set per epoch history = LossHistory() # initialise Metrics used for calculating NDCG@k per epoch # and saving the model weights with the highest NDCG@k value metrics = Metrics( model=self.model, val_tr=x_val_tr, val_te=x_val_te, mapper=mapper, k=self.k, save_path=self.save_path, ) self.reduce_lr = ReduceLROnPlateau( monitor="val_loss", factor=0.2, patience=1, min_lr=0.0001 ) if self.annealing: # initialise AnnealingCallback for annealing process anneal = AnnealingCallback( self.beta, self.anneal_cap, self.total_anneal_steps ) # fit model self.model.fit_generator( generator=self.nn_batch_generator(x_train), steps_per_epoch=self.number_of_batches, epochs=self.n_epochs, verbose=self.verbose, callbacks=[metrics, history, self.reduce_lr, anneal], validation_data=(x_valid, x_valid), ) self.ls_beta = anneal.get_data() else: self.model.fit_generator( generator=self.nn_batch_generator(x_train), steps_per_epoch=self.number_of_batches, epochs=self.n_epochs, verbose=self.verbose, callbacks=[metrics, history, self.reduce_lr], validation_data=(x_valid, x_valid), ) # save lists self.train_loss = history.losses self.val_loss = history.val_losses self.val_ndcg = metrics.get_data() def get_optimal_beta(self): - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - recommenders/models/vae/standard_vae.py [310:416]: - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - def _take_sample(self, args): """Sample epsilon ∼ N (0,I) and compute z via reparametrization trick.""" """Calculate latent vector using the reparametrization trick. The idea is that sampling from N (_mean, _var) is s the same as sampling from _mean+ epsilon * _var where epsilon ∼ N(0,I).""" # sampling from latent dimension for decoder/generative part of network _mean, _log_var = args epsilon = K.random_normal( shape=(K.shape(_mean)[0], self.latent_dim), mean=0.0, stddev=1.0, seed=self.seed, ) return _mean + K.exp(_log_var / 2) * epsilon def nn_batch_generator(self, x_train): """Used for splitting dataset in batches. Args: x_train (numpy.ndarray): The click matrix for the train set with float values. """ # Shuffle the batch np.random.seed(self.seed) shuffle_index = np.arange(np.shape(x_train)[0]) np.random.shuffle(shuffle_index) x = x_train[shuffle_index, :] y = x_train[shuffle_index, :] # Iterate until making a full epoch counter = 0 while 1: index_batch = shuffle_index[ self.batch_size * counter : self.batch_size * (counter + 1) ] # Decompress batch x_batch = x[index_batch, :] y_batch = y[index_batch, :] counter += 1 yield (np.array(x_batch), np.array(y_batch)) # Stopping rule if counter >= self.number_of_batches: counter = 0 def fit(self, x_train, x_valid, x_val_tr, x_val_te, mapper): """Fit model with the train sets and validate on the validation set. Args: x_train (numpy.ndarray): The click matrix for the train set. x_valid (numpy.ndarray): The click matrix for the validation set. x_val_tr (numpy.ndarray): The click matrix for the validation set training part. x_val_te (numpy.ndarray): The click matrix for the validation set testing part. mapper (object): The mapper for converting click matrix to dataframe. It can be AffinityMatrix. """ # initialise LossHistory used for saving loss of validation and train set per epoch history = LossHistory() # initialise Metrics used for calculating NDCG@k per epoch # and saving the model weights with the highest NDCG@k value metrics = Metrics( model=self.model, val_tr=x_val_tr, val_te=x_val_te, mapper=mapper, k=self.k, save_path=self.save_path, ) self.reduce_lr = ReduceLROnPlateau( monitor="val_loss", factor=0.2, patience=1, min_lr=0.0001 ) if self.annealing: # initialise AnnealingCallback for annealing process anneal = AnnealingCallback( self.beta, self.anneal_cap, self.total_anneal_steps ) # fit model self.model.fit_generator( generator=self.nn_batch_generator(x_train), steps_per_epoch=self.number_of_batches, epochs=self.n_epochs, verbose=self.verbose, callbacks=[metrics, history, self.reduce_lr, anneal], validation_data=(x_valid, x_valid), ) self.ls_beta = anneal.get_data() else: self.model.fit_generator( generator=self.nn_batch_generator(x_train), steps_per_epoch=self.number_of_batches, epochs=self.n_epochs, verbose=self.verbose, callbacks=[metrics, history, self.reduce_lr], validation_data=(x_valid, x_valid), ) # save lists self.train_loss = history.losses self.val_loss = history.val_losses self.val_ndcg = metrics.get_data() def get_optimal_beta(self): - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -