ardendertat / applied-deep-learning-with-keras Goto Github PK

Can you add a readme file with the posts you made in medium ???

Minor correction in function plot_loss_accuracy(history) needed.
Run this on Keras 2.3.1 with correction from
acc = history.history['acc'][-1] to
acc = history.history['accuracy'][-1]

Note that
history.history.keys()
gives:
dict_keys(['loss','accuracy'])

Running the notebook on JupyterLab 1.2.4 / python 3.6 requires from __future__ import print_function in the first line of the first cell.

Great tutorial and notebooks! Thanks.

I think the model variables in the multiclass classification visualization section need to be updated from just "model" to "sr_model" (first snippet) and "deep_model" (second snippet) below:

y_pred_class = model.predict_classes(X_test, verbose=0)
y_test_class = np.argmax(y_test, axis=1)
print(classification_report(y_test_class, y_pred_class))
plot_confusion_matrix(model, X_test, y_test_class)

y_pred_class = model.predict_classes(X_test, verbose=0)
y_test_class = np.argmax(y_test, axis=1)
print(classification_report(y_test_class, y_pred_class))
plot_confusion_matrix(model, X_test, y_test_class)

Thanks for the great tutorials!

Just a small suggestion, in Part 3: Autoencoders, the function may be cleaned up further.

def plot_autoencoder_outputs(autoencoder, n, dims):
    decoded_imgs = autoencoder.predict(x_test)

    # number of example digits to show
    n = 5  # Should be commented out otherwise n is overwritten. 

`X, y = make_sine_wave()

plot_data(X, y, figsize=(10, 8))`

Generates a plot that appears to be all red 0 with 6 blue does (as 1).