def main():
(X_train, y_train), (X_test, y_test) = tf.keras.datasets.mnist.load_data()
X_train = X_train.astype(np.float32).reshape((60000, 28, 28, 1)) / 255.0
X_test = X_test.astype(np.float32).reshape((10000, 28, 28, 1)) / 255.0
model = create_model()
loss = tf.keras.losses.SparseCategoricalCrossentropy()
acc = tf.keras.metrics.SparseCategoricalAccuracy()
optim = tf.keras.optimizers.Adam()
# train
model.compile(optimizer=optim, loss=loss, metrics=[acc])
model.fit(X_train, y_train, validation_data=(X_test, y_test), epochs=1, batch_size=2048)
val_loss, val_acc = model.evaluate(X_test, y_test, batch_size=2048)
parser = tfcg.from_graph_def(tf.get_default_graph().as_graph_def())
parser.dump_img("outputs/mnist_graph.png")
parser.dump_yml("outputs/mnist_graph.yml")
print(val_loss, val_acc)
x = tf.keras.layers.Conv2D(16, 3, input_shape=[28, 28, 3])(x)
x = tf.keras.layers.Conv2D(32, 1)(x)
x = tf.keras.layers.Conv2D(64, 2)(x)
x = tf.keras.layers.Conv2D(128, 2)(x)
x = tf.keras.layers.Flatten()(x)
x1 = tf.keras.layers.Dense(32)(x)
x1 = tf.keras.layers.ReLU()(x1)
x1 = tf.keras.layers.Dense(16)(x1)
x2 = tf.keras.layers.Dense(128)(x)
x2 = tf.keras.layers.ReLU()(x2)
x2 = tf.keras.layers.Dense(64)(x2)
x2 = tf.keras.layers.ReLU()(x2)
x2 = tf.keras.layers.Dense(32)(x2)
x2 = tf.keras.layers.ReLU()(x2)
x2 = tf.keras.layers.Dense(16)(x2)
return x, x2
with tf.Graph().as_default() as graph:
x = np.random.rand(128, 28, 28, 3)
x_p = tf.placeholder(tf.float32, [None, 28, 28, 3])
out1, out2 = build(x_p)
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
o = sess.run((out1, out2), feed_dict={x_p: x})
parser = tfcg.from_graph_def(sess.graph_def)
parser.dump_img('outputs/multitask_graph.png')
parser.dump_yml('outputs/multitask_graph.yml')
parser.dump_gml('outputs/multitask_graph.gml')