Auto Encoder Experiments

How to run it?

Example Usage:

python auto_encoder_gd.py --input_dim 32 --nb_fixed_point 2  --constant 1 --nb_layer 2 --hidden_dim 1000 --act sigmoid

Please use python auto_encoder_gd.py -h to get info on what those parameters are.

Output

The program will print the following data:

input_dim, constant, nb_fixed_point, nb_layer, hidden_dim, average iteration to converge, average initial loss, average final loss, average initial Jacobian norm, final Jacobian norm, the average difference between initial and final Jacobian norm

It will also save all the output data in NumPy arrays, including the Jacobina matrices of networks. Please refer to the code to get details on that.