hist_loss.append(h.history['loss'])
hist_acc = np.hstack(hist_acc)
hist_loss = np.hstack(hist_loss)
plt.figure()
plt.plot(hist_acc)
plt.legend(['train', 'val'])
plt.xlabel('Epochs')
plt.ylabel('Accuracy')
plt.figure()
plt.plot(hist_loss)
plt.legend(['train', 'val'])
plt.xlabel('Epochs')
plt.ylabel('Loss')
#%% Performance on training dataset
y_hat_train = mdl.predict(X_train)
y_pred_train = np.argmax(y_hat_train, axis=1)
acc_train = accuracy_score(y_train, y_pred_train)
cm_train = confusion_matrix(y_train, y_pred_train)
names = ['berry', 'bird', 'dog', 'flower', 'other']
plt.figure()
sns.heatmap(cm_train.astype(int),
annot=True,
fmt='d',
cmap="YlGnBu",
xticklabels=names,
yticklabels=names)
#%% Loading test data
batch_size=10,
validation_data=({
'model_input': x_test
}, {
'd_output': x_test,
'p_output': y_test
}))
encoder = Model(main_input, encoded, name="encoder")
decoded_input = Input(shape=(4, ))
decoded_p = full_model.get_layer('d1')(decoded_input)
decoded_p = full_model.get_layer('d2')(decoded_p)
decoded_p = full_model.get_layer('d_output')(decoded_p)
decoder = Model(decoded_input, decoded_p)
regression = Model(main_input, regression)
encoded_res = encoder.predict(x_test)
decoded_res = decoder.predict(encoded_res)
regression_res = regression.predict(x_test)
decoder.save('decoder.h5')
encoder.save('encoder.h5')
regression.save('regression.h5')
pd.DataFrame(np.round(x_test, 5)).to_csv("x_test.csv")
pd.DataFrame(np.round(y_test, 5)).to_csv("y_test.csv")
pd.DataFrame(np.round(x_train, 5)).to_csv("x_train.csv")
pd.DataFrame(np.round(y_train, 5)).to_csv("y_train.csv")
pd.DataFrame(np.round(encoded_res, 5)).to_csv("encoded_res.csv")
pd.DataFrame(np.round(decoded_res, 5)).to_csv("decoded_res.csv")
pd.DataFrame(np.round(regression_res, 5)).to_csv("regression_res.csv")
