How to use PENetwork for inference?

[thatdc]

As the title says, I'm trying to use PENetwork for inference with PE files of my choice.
I extracted the features with the code found on the github of Ember2018 (features version 2) and fed the 2381 long tensor to the PENetwork, with the weights of seed0-epoch10.

I always obtain a very high value off the 'malware' head (~0.99) both when I use a malware PE or a goodware PE.
Am I missing something? Do I have to preprocess the ember features?

I'd troubleshoot by myself, but I'm having a hard time finding the documentation on how a PE is handled before submitting it to the PENetwork.