iteration = 15000 batch_size = 1000 file_name='warehouse' fileObject = open(file_name,'rb') years = pickle.load(fileObject) timbres = pickle.load(fileObject) pitches = pickle.load(fileObject) min_length = pickle.load(fileObject) non_time_features = pickle.load(fileObject) fileObject.close() #train_timbres,cv_timbres,test_timbres = songProcess.dataset_split(timbres) #train_pitches,cv_pitches,test_pitches = songProcess.dataset_split(pitches) train_non_time,cv_non_time,test_non_time = songProcess.dataset_split(non_time_features) train_year,cv_year,test_year = songProcess.dataset_split(years) #sklearn-begin ''' cv_array = np.arange(0, len(cv_year)) cv_X_without_year,cv_y = songProcess.get_batch_non_time_data(cv_array,cv_non_time,cv_year) cv_y = songProcess.transfer_year_to_decade(cv_y) index_array = np.arange(0, np.size(train_year)) batch_X_without_year,batch_y = songProcess.get_batch_non_time_data(index_array,train_non_time,train_year) batch_y_10d = songProcess.transfer_year_to_10d(songProcess.transfer_year_to_decade(batch_y)) scaler = MinMaxScaler(feature_range=(-5,5)) scaler.fit(batch_X_without_year) batch_X_without_year = scaler.transform(batch_X_without_year)
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
iteration = 10000
batch_size = 256
file_name = 'warehouse'
fileObject = open(file_name, 'rb')
years = pickle.load(fileObject)
timbres = pickle.load(fileObject)
pitches = pickle.load(fileObject)
min_length = pickle.load(fileObject)
non_time_features = pickle.load(fileObject)
fileObject.close()
train_timbres, cv_timbres, test_timbres = songProcess.dataset_split(timbres)
train_pitches, cv_pitches, test_pitches = songProcess.dataset_split(pitches)
train_non_time, cv_non_time, test_non_time = songProcess.dataset_split(
non_time_features)
train_year, cv_year, test_year = songProcess.dataset_split(years)
def add_layer(inputs, in_size, out_size, activation_function=None):
Weights = tf.Variable(tf.random_normal([in_size, out_size]))
biases = tf.Variable(tf.zeros([1, out_size]) + 0.1)
Wx_plus_b = tf.matmul(inputs, Weights) + biases
if activation_function is None:
outputs = Wx_plus_b
else:
outputs = activation_function(Wx_plus_b)
return outputs
batch_size = 64
'''Data processing'''
#read data
file_name='warehouse'
fileObject = open(file_name,'rb')
years = pickle.load(fileObject)
timbres = pickle.load(fileObject)
pitches = pickle.load(fileObject)
min_length = pickle.load(fileObject)
fileObject.close()
total_size = len(years)
'''Data Spliting'''
#training set
train_timbres,cv_timbres,test_timbres = songProcess.dataset_split(timbres)
train_pitches,cv_pitches,test_pitches = songProcess.dataset_split(pitches)
train_years,cv_years,test_years = songProcess.dataset_split(years)
cv_array = np.arange(0, len(cv_years))
cv_X, cv_y = songProcess.get_batch_data(cv_array, min_length, cv_timbres, cv_pitches, cv_years)
'''RNN Model Definition'''
# tensorflow placeholders
tf_x = tf.placeholder(tf.float32, [None, min_length * INPUT_SIZE]) # shape(batch, 784) [time_steps, batch_size, num_features]
image = tf.reshape(tf_x, [-1, INPUT_SIZE,min_length]) # (batch, height, width, channel)
tf_y = tf.placeholder(tf.int32, [None, 10]) # input y
# RNN
rnn_cell = tf.contrib.rnn.BasicLSTMCell(
num_units=300,
forget_bias=0.8,