res = sess.run(
[tf.nn.softmax(preds["n0"]),
tf.nn.softmax(preds["n1"])],
feed_dict={
placeholders["i0"]: inputs["i0"],
placeholders["i1"]: inputs["i1"]
})
sess.close()
# Return both accuracies
return accuracy_error(res[0], outputs["o0"]), accuracy_error(
res[1], outputs["o1"])
if __name__ == "__main__":
fashion_x_train, fashion_y_train, fashion_x_test, fashion_y_test = load_fashion(
)
mnist_x_train, mnist_y_train, mnist_x_test, mnist_y_test = load_mnist()
OHEnc = OneHotEncoder(categories='auto')
fashion_y_train = OHEnc.fit_transform(np.reshape(fashion_y_train,
(-1, 1))).toarray()
fashion_y_test = OHEnc.fit_transform(np.reshape(fashion_y_test,
(-1, 1))).toarray()
mnist_y_train = OHEnc.fit_transform(np.reshape(mnist_y_train,
(-1, 1))).toarray()
mnist_y_test = OHEnc.fit_transform(np.reshape(mnist_y_test,
(-1, 1))).toarray()
return predictions
def eval_cnn_ae(preds, placeholders, sess, graph, inputs, _, __):
with graph.as_default():
res = sess.run(preds["n1"], feed_dict={placeholders["i0"]: inputs["i0"]})
sess.close()
if np.isnan(res).any():
return 288,
else:
return mean_squared_error(np.reshape(res, (-1)), np.reshape(inputs["i0"], (-1))),
if __name__ == "__main__":
x_train, y_train, x_test, y_test = load_fashion()
x_train = np.expand_dims(x_train, axis=3)/255
x_train = np.concatenate((x_train, x_train, x_train), axis=3)
x_test = np.expand_dims(x_test, axis=3)/255
x_test = np.concatenate((x_test, x_test, x_test), axis=3)
OHEnc = OneHotEncoder(categories='auto')
y_train = OHEnc.fit_transform(np.reshape(y_train, (-1, 1))).toarray()
y_test = OHEnc.fit_transform(np.reshape(y_test, (-1, 1))).toarray()
# Here we define a convolutional-transposed convolutional network combination
e = Evolving(loss=train_cnn_ae, desc_list=[ConvDescriptor, TConvDescriptor], x_trains=[x_train], y_trains=[y_train], x_tests=[x_test], y_tests=[y_test], evaluation=eval_cnn_ae, batch_size=150, population=2, generations=10, n_inputs=[[28, 28, 3], [7, 7, 1]], n_outputs=[[49], [28, 28, 3]], cxp=0, mtp=1, hyperparameters = {"lrate": [0.1, 0.5, 1], "optimizer": [0, 1, 2]}, no_batch_norm=False, no_dropout=False)
a = e.evolve()
predictions = np.sort(predictions, axis=1)
predictions = np.mean(predictions, axis=0)
return -np.sum([
aux_preds[w] *
np.log(aux_preds[w] / predictions[w]) if aux_preds[w] > 0 else 0
for w in range(predictions.shape[0])
]),
if __name__ == "__main__":
mobile_graph, model = load_model(
) # The model and its graph are used as global variables
x_train, _, x_test, _ = load_fashion()
# The GAN evolutive process is a common 2-DNN evolution
e = Evolving(loss=gan_train,
desc_list=[MLPDescriptor, MLPDescriptor],
x_trains=[x_train],
y_trains=[x_train],
x_tests=[x_test],
y_tests=[x_test],
evaluation=gan_eval,
batch_size=150,
population=10,
generations=10,
n_inputs=[[28, 28], [10]],
n_outputs=[[1], [784]],
cxp=0.5,
mtp=0.5)