def __init__(self, train):
dataset_number = env.DATASET_NUMBER
num_dataset_to_use = env.NUM_DATASET_TO_USE
chords_in_vector = env.CHORDS_IN_BAR * dataset_number
chord_length = env.CHORD_LENGTH
num_layers = env.NUM_LAYERS
num_hidden = env.NUM_HIDDEN
batch_size = env.BATCH_SIZE
epoch = env.EPOCH
dropout_pb = env.DROPOUT_PB
with tf.variable_scope(str(dataset_number) + str(num_dataset_to_use) + str(chords_in_vector) + str(chord_length)):
cell_type = tf.nn.rnn_cell.GRUCell
# Number of examples, number of input, dimension of each input
data = tf.placeholder(tf.float64, [None, chords_in_vector, chord_length])
target = tf.placeholder(tf.float64, [None, 2])
cell = cell_type(num_hidden)
reader = Reader()
if train:
reader.read_training_data(dataset_number, num_dataset_to_use)
reader.read_testing_data(dataset_number, num_dataset_to_use)
# make sure it has the correct format for the RNN
reader.convert_to_rnn_format(chords_in_vector, chord_length)
self.model = model = MultiRNNModel(cell, data, target, train, batch_size, epoch, dropout_pb, num_hidden, num_layers, reader.training_attributes, reader.training_labels, reader.testing_attributes, reader.testing_labels)
else:
#reader.read_training_data(dataset_number, 0)
#reader.read_testing_data(dataset_number, num_dataset_to_use)
# make sure it has the correct format for the RNN
#reader.convert_to_rnn_format(chords_in_vector, chord_length)
self.model = model = MultiRNNModel(cell, data, target, train, batch_size, epoch, dropout_pb, num_hidden, num_layers)
from pprint import pprint np.set_printoptions(precision=6, suppress=True) # GRU dataset_number = 8 num_dataset_to_use = 1000 chords_in_vector = 4 * dataset_number chord_length = 33 reader = Reader() reader.read_training_data(dataset_number, num_dataset_to_use) reader.read_testing_data(dataset_number, num_dataset_to_use) # make sure it has the correct format for the RNN reader.convert_to_rnn_format(chords_in_vector, chord_length) cell_index = 0 batch_size = 1000 num_hidden = 60 num_layers = 10 epoch = 67 # Number of examples, number of input, dimension of each input data = tf.placeholder(tf.float64, [None, chords_in_vector, chord_length]) target = tf.placeholder(tf.float64, [None, 2]) model = MultiRNNModel(data, target, 0.1, num_hidden=num_hidden, num_layers=num_layers)
