DeepRank output format

[manonreau]

Describe the bug DeepRank output format is not user friendly

Environment: Not relevant

To Reproduce See any example from https://github.com/DeepRank/deeprank/tree/master/paper_pretrained_models

Expected Results The output of the network is an HDF5 file that contains the raw network prediction (no softmax transformation for classification tasks) which is hard to understand for someone unfamiliar with NN. Extracting a score or a rank from the default output is not straightforward.

It would be nice to add:

• for classification task: apply a softmax transformation, and eventually save the raw data, the transformed ones, and the predicted class
• a hdf5-to-csv convertor to easily extract the information from the hdf5 output files and provide a human-readable format

This is already implemented in DeepRank-GNN and the hdf5 to csv convertor can be easily adapted (see https://github.com/DeepRank/Deeprank-GNN/blob/master/deeprank_gnn/tools/hdf5_to_csv.py)

Actual Results or Error Info

Additional Context