# print "X shape : ",X.get_shape() # i+=1 x = np.reshape(x, (words_to_read, embedding_size)) # print x.shape return x, y # nX = len(X) # print "DATA VECTORIZED............." # X = np.reshape(X,(nX,words_to_read,1)) # print X.get_shape(),Y.get_shape() sess = tf.Session() K.set_session(sess) model = Sequential() model.add( LSTM(256, input_shape=(int(X.get_shape()[1]), int(X.get_shape()[2])), return_sequences=True)) model.add(Dropout(0.2)) model.add(LSTM(256, return_sequences=False)) model.add(Dropout(0.2)) model.add(Dense(vocab_size)) #, activation="softmax")) # model.compile(loss="categorical_crossentropy",optimizer="adam") # model = LSTM(256,input_shape=(X.get_shape()[1],X.get_shape()[2]),init='uniform',return_sequences=True)(X) # model = Dropout(0.2)(model) # model = LSTM(256)(model)
def ini_chainer(): chainer.cuda.get_device(chainer_ID).use() google_net.to_gpu(chainer_ID) paintschainer.to_gpu(chainer_ID) print('chainer initialized') chainer_thread = threading.Thread(target=ini_chainer) chainer_thread.start() session = tf.Session(config=tf.ConfigProto(gpu_options=tf.GPUOptions( visible_device_list=str(tensorflow_GPU_ID), per_process_gpu_memory_fraction=k_between_tf_and_chainer))) K.set_session(session) EPS = 1e-12 lr = 1e-6 beta1 = 0.5 with tf.variable_scope("generator"): base_generator = load_model('base_generator.net') sketch_ref_input_448 = tf.placeholder(dtype=tf.float32, shape=(None, None, None, 1)) local_hint_input_448 = tf.placeholder(dtype=tf.float32, shape=(None, None, None, 3)) hint_s57c64_0 = tf.placeholder(dtype=tf.float32, shape=(None, 64)) hint_s29c192_0 = tf.placeholder(dtype=tf.float32, shape=(None, 192)) hint_s29c256_0 = tf.placeholder(dtype=tf.float32, shape=(None, 256)) hint_s29c320_0 = tf.placeholder(dtype=tf.float32, shape=(None, 320))