def examples(self, data_generator, num_examples=5): """ Prints some examples during training Args: data_generator: """ assert self.symbols x, y = data_generator.next_batch(validation=True) # input_strings = decode_output_sequences(x, symbols=SYMBOLS) target_strings = decode_output_sequences(y, symbols=self.symbols) model_output = self.predict(x) pred_strings = decode_output_sequences(model_output, symbols=self.symbols) print(target_strings[:num_examples]) print(pred_strings[:num_examples])
logging = tf.logging flags.DEFINE_string( "model", "small", "A type of model. Possible options are: small, medium, large, zoneout.") flags.DEFINE_string("data_path", None, "data_path") FLAGS = flags.FLAGS from program_generator import ProgramGenerator, SYMBOLS, SYMBOL_TO_IDX, INPUT_SEQ_LEN, OUTPUT_SEQ_LEN program_generator = ProgramGenerator(batch_size=10, program_length=1, num_len=2) x, y = program_generator.next_batch() input_strings = decode_output_sequences(x, symbols=SYMBOLS) target_strings = decode_output_sequences(y, symbols=SYMBOLS) print(" Inputs:", input_strings) print("Targets:", target_strings) session.close() tf.reset_default_graph() session = tf.InteractiveSession() hidden_units = 320 num_layers = 2 training_batch_size = 128 num_symbols = len(SYMBOL_TO_IDX) program_model = Seq2SeqProgramModel(session=session, hidden_units=hidden_units,
num_layers = 2 training_batch_size = 32 z_prob_cells = 0.05 z_prob_states = 0 num_symbols = len(SYMBOL_TO_IDX) DEFAULT_LEARNING_RATE = 0.01 print('zoneout cells'+str(z_prob_cells)+'states'+str(z_prob_states)) ################# # DATA ################# addition_generator = AdditionGenerator(batch_size=3) x, y = addition_generator.next_batch() input_strings = decode_output_sequences(x, symbols=SYMBOLS) target_strings = decode_output_sequences(y, symbols=SYMBOLS) print(" Inputs:", input_strings) print("Targets:", target_strings) session.close() tf.reset_default_graph() session = tf.InteractiveSession() ################# # MODEL ################# # Wrapper for the TF RNN cell # For an LSTM, the 'cell' is a tuple containing state and cell # We use TF's dropout to implement zoneout