search

rjpg / bftensor

tensorflow on time-series prediction
13 stars 5 forks source link

Is AutoEncoderMNIST.py a stacked autoencoder? #2

Open josuerocha opened 5 years ago

josuerocha commented 5 years ago

Dear rjpg,

Is AutoEncoderMNIST.py a stacked autoencoder? Could you please provide clarification on that?

Thanks.

rjpg commented 5 years ago

Hello,

yes, it is a test for the principle of use autoenconder for classification.

It is using dense layers in MNIST ... optimally I should use CNN2D autoencoder when treating images but here the goal was just to put the principle in execution (using MNIST is just a rapid way to have some dataset ).

basically, we have 3 main models : Auto encoder (encoder and decoder together) decoder alone *and added model in front of the encoder for fine tunning and classification.

The goal of the autoencoder is to reduce features and then use this features, provided by the encoder (1st half of the autoencoder), to apply the classification process (add more layers in front of the encoder for fine tunning and classification)