Code developed by Tristan during his 2022 M2 intership in Mnemosyne team.
pandas 1.3.4 tensorflow 2.8.0 keras 2.8.0 ikpy 3.3
- 2 preprocessing files for respectively the emulator and controller named preprocessing_emulator.py and preprocessing_controller.py
- 2 files with tools for the preprocessing named remap_utils.py and rot_quat_utils.py
- 1 file for various plots named plot.py
- 2 files for the neural networks architectures named controller.py and emulator.py
- 1 file for the hyperparameter search of the emulator named hyperparam_search.py
- 1 file for hyperparameter plotting named _hyperplot.py (a small modification by Nathan of reservoirpy Matplotlib wrapper https://github.com/reservoirpy/reservoirpy/blob/master/reservoirpy/hyper/_hyperplot.py)
- copy file Index des sujets.json in datasets folder (needed for preprocessing_emulator.py and preprocessing_controller.py)
- create folder datasets and folder saved_models
- create subfolder dataset/training_data
- run preprocessing_emulator.py : it will prepare and save the training dataset in folder datasets/training_data
- run emulator.py : it will train the network and save the model and its history of training in folder saved_models
- Do # steps in order to train the emulator (similar files required for preprocessing_controller)
- run preprocessing_controller.py : it will prepare and save the training dataset in folder datasets/training_data
- run controller.py : it will train the network and save the model and its history of training in folder saved_models
- in emulator.py or controller.py it's better to run the file in a spyder console to have a direct visu of the results
- you can run emulator.py and controller.py cell by cell at each ##UNCOMMENT TO RUN FILE tags
- for lines or functions preceded by ##UNCOMMENT IF, read the associated comment
- some functions in emulator.py can be imported and used for the controller such as load_model, load_hist
- code main functions for controller, and hyparam_search