neural_prediction

This folder contains the code to use pre-trained neural networks to predict experimental monkey neural data:

The folder contains the code to predict neural data using neural network models trained on a task. The Code contains the following files, to execute in this order:

• 1_run_sessions_75pca_h5_align.sh & 1_run_sessions_75pca_h5_align_passive.sh bash script to generate network's activations giving as input the test behavioral muscle spindles inputs to the frozen neural networks (run in the docker). The PCs of the activations are stored. The same can be applied for untrained models by adding the _untrained suffix. It uses the following scripts generate_session_activations_active.py & generate_session_activations_passive.py.
• 2_run_predictions_75pca_h5_align.sh & 2_run_predictions_75pca_h5_align_passive.sh bash script to predict neural activity using the previously generated network's activations (use conda environment DeepProprio). The same can be applied for untrained models by adding the _untrained suffix. It uses the following scripts compute_session_predictivity_h5_cv_pool_align.py, compute_model_session_predictivity_h5_cv_align.py & compute_session_predictivity_h5_cv_pool_align_passive.py, compute_model_session_predictivity_h5_cv_align_passive.py
• 3_fit_linear_models_cv.sh & 3_fit_linear_models_cv_passive.sh bash script to predict neural activity with linear models using task-related variables(use conda environment DeepProprio).

To run the same scripts but using passive data, add '_passive' as suffix to the script name (e.g. run_sessions_75pca_h5_align.sh --> run_sessions_75pca_h5_align_passive.sh)