Python : Save and Load Tensorflow Keras Model

This post shows how to save and load your Keras model. The loaded model contains the same parameters (weights and biases) and the model structure as the final model you saved.

Save and Load Tensorflow Keras Model

In this post, we use the same data in the following previous post so I omit the data preparation part.

• Python : Tensorflow Keras Tuner for the optimization of hyper parameters

Fitting a Keras model and Saving it as h5 file

As a simple example, the following deep learning model is constructed. After fitting the model and forecasting using the test data, this Keras deep learning model is saved by using save() function. The saved file name has the h5 extention.

from keras.models import Sequential from keras.layers import Dense, LSTM, Dropout from keras.callbacks import EarlyStopping from keras import optimizers   #============================================== # define a simple Sequential model #============================================== seed_everything(1) # fix the random seed model = Sequential() model.add(LSTM(50, input_shape=(n_steps,1))) model.add(Dropout(0.2)) model.add(Dense(6)) model.add(Dense(1)) model.compile(loss='mean_squared_error',                optimizer='adam',                metrics = ['mse'])   #============================================== # fit the model #============================================== early_stop = EarlyStopping(monitor='loss',                    patience=50, verbose=0)   model.fit(X_train, y_train, epochs=1000,            batch_size=32, verbose=2,            callbacks=[early_stop])   #============================================== # Evaluate model #============================================== test_loss, test_acc = model.evaluate(X_test,  y_test, verbose=2) print(test_loss, test_acc)   #============================================== # Forecast using test data #============================================== y_pred = model.predict(X_test, verbose=0)   plt.figure().set_figwidth(12) plt.plot(np.c_[y_pred, y_test]) plt.legend(('Test data','Forecast')) plt.show()   #============================================== # Save the final model #============================================== model.save('model_1234.h5')     Colored by Color Scripter

cs

Loading the Keras model

Loading the Keras model is straightforward; simply use the load_model() function with the saved file name and the h5 file extension.

#============================================== # Load the final model saved #============================================== model_loaded = tf.keras.models.load_model('model_1234.h5')   #============================================== # Evaluate model #============================================== test_loss, test_acc = model_loaded.evaluate(X_test, y_test, verbose=2) print(test_loss, test_acc)   #============================================== # Forecast using test data #============================================== y_pred_from_model_loaded = model.predict(X_test, verbose=0)   plt.figure().set_figwidth(12) plt.plot(np.c_[y_pred_from_model_loaded, y_test]) plt.legend(('Test data','Forecast from the model loaded')) plt.show()     Colored by Color Scripter

cs

The model saved is located in the working directory as follows.

Concluding Remarks

This post explained how to save and load your Keras model. This appears to be beneficial for preserving the reproducibility of your study. \(\blacksquare\)