Autoencoder is not predicting for some nodes in the botteneck layer

Question

I\'m using tf.Keras to build a deep-fully connected autoencoder. My input dataset is a dataframe with shape (19947,), and the purpose of the autoencoder is to predict normal