def build_model():
utils.set_nn_config()
inp = Input(config.MODULE_INPUT_DIM)
layer = inp
layer = utils.conv_layer(layer, CONV_FILTERS, KERNEL_SIZE,
REGULARIZER_CONST)
for _ in range(CONV_LAYERS):
layer = utils.conv_layer(layer, CONV_FILTERS, KERNEL_SIZE,
REGULARIZER_CONST)
out = output_layer(layer)
model = Model(inputs=inp, outputs=out)
compile_model(model)
return model
def build_model():
utils.set_nn_config()
inp = Input(config.SYMBOLS_INPUT_DIM)
layer = utils.conv_layer(inp, CONV_FILTERS, KERNEL_SIZE, REGULARIZER_CONST)
layer = utils.conv_layer(layer, CONV_FILTERS*2, KERNEL_SIZE, REGULARIZER_CONST)
layer = Dropout(0.25)(layer)
layer = MaxPooling2D(pool_size=(2, 2), padding="same")(layer)
layer = Dense(DENSE_UNITS, activation='relu')(layer)
layer = MaxPooling2D(pool_size=(2, 2), padding="same")(layer)
layer = Flatten()(layer)
layer = Dense(DENSE_UNITS // 2, activation='relu')(layer)
layer = Dropout(0.5)(layer)
out = Dense(config.SYMBOLS_OUTPUT_DIM, activation='softmax')(layer)
model = Model(inputs=inp, outputs=out)
compile_model(model)
return model
def build_model():
sess = utils.get_nn_config()
graph = tf.Graph()
inp = Input(config.MODULE_INPUT_DIM)
layer = inp
layer = utils.conv_layer(layer, CONV_FILTERS, KERNEL_SIZE,
REGULARIZER_CONST)
for i in range(CONV_LAYERS):
layer = utils.conv_layer(layer, CONV_FILTERS, KERNEL_SIZE,
REGULARIZER_CONST)
#if i % 3 == 0:
# layer = Dropout(0.3)(layer)
out = output_layer(layer)
model = Model(inputs=inp, outputs=out)
compile_model(model)
model._make_predict_function()
return model, graph, sess
def build_model():
sess = utils.get_nn_config()
graph = Graph()
inp = Input(config.SYMBOLS_INPUT_DIM)
layer = utils.conv_layer(inp, CONV_FILTERS, KERNEL_SIZE, REGULARIZER_CONST)
layer = utils.conv_layer(layer, CONV_FILTERS * 2, KERNEL_SIZE,
REGULARIZER_CONST)
layer = Dropout(0.25)(layer)
layer = MaxPooling2D(pool_size=(2, 2), padding="same")(layer)
layer = Dense(512, activation='relu')(layer)
layer = MaxPooling2D(pool_size=(2, 2), padding="same")(layer)
layer = Flatten()(layer)
layer = Dense(256, activation='relu')(layer)
layer = Dropout(0.5)(layer)
out = Dense(config.SYMBOLS_OUTPUT_DIM, activation='softmax')(layer)
model = Model(inputs=inp, outputs=out)
compile_model(model)
model._make_predict_function()
return model, graph, sess