Code and data accompanying our SVRHM'21 paper.
Requires tensorflow 1.13, python 3.7, scikit-learn, and pytorch 1.6.0 to be installed.
Python scripts included:
- RNN_gen.py can be used to train the RNNs.
- RNN_analyse_reprs_recurrence.py can be used to train linear classifiers for auxiliary variables and category decoding, on the layer activations and recurrent flows
- RNN_perturb.py can be used to perform the perturbation analyses.
Two pretrained RNNs can be downloaded from https://osf.io/pf4u5/ to be assessed using the jupyter notebook 'SVRHM21_results.ipynb'. Alternatively, you can train your own RNN and assess it.