######################################################
x, y = pickle.load(open('train.pkl', 'rb'))

x_hist = np.zeros((x.shape[0], 36))
for i, row in enumerate(x):
    hist, _ = np.histogram(row, bins=36)
    x_hist[i] = hist

######################################################
x_train = x_hist[0:10000]
y_train = y[0:10000]

x_val = x_hist[10000:12500]
y_val = y[10000:12500]
######################################################

mlp = MLP(n_features=x_train.shape[1], n_hidden=50, n_output=10,
          eta=0.001, decrease_const=0.000005, l1=0.1, l2=0.1, alpha=0.001, batch_size=50)

costs = mlp.fit(x_train, y_train, epochs=1000, show_progress=True)

# validation
predictions = mlp.predict(x_val)
accuracy = np.sum(y_val == predictions, axis=0) / x_val.shape[0]
print("Validation accuracy: %.2f" % accuracy)
######################################################