search

jeyabbalas / tabnet

A TensorFlow 2 Keras implementation of TabNets.
MIT License
13 stars 1 forks source link

Unsupervised pre-training? #1

Open elbamos opened 1 year ago

elbamos commented 1 year ago

Do you have an example showing how to do unsupervised pre training with this code?

julianbayerl commented 4 months ago

The problem is this line 

       ` x = self.step_feature_transformers[step](inputs, 
                                                 training=False)
        x = self.step_fc_layers[step](inputs)
        reconstructed_features += x`

in the decoder. Also note that (at least for me) batch_size is not input_shape[-1], so if you correct for these two things it should work.