Autoencoders with missing values (Keras, Python)

Question

I am trying to build a simple linear Autoencoder model for the following dataset:

permno         10001     10002     10012  ...     93434  93435     93436
dat