What matters in reinforcement learning for tractography - Trained models

Trained models for "What matters in reinforcement learning for tractography". These can be loaded to track on arbitrary data.

Nomenclature goes as follows:

[RL algorithm]_Train[Dataset][Extra]Exp[1-5]

RL algorithm can be one of the following: VPG, A2C, ACKTR, TRPO, PPO, DDPG, TD3, SAC, SAC_Auto
Dataset refers to the dataset used for training, can be either FiberCup or ISMRM2015
Exp1 to Exp2 refer to the experiment the agent were trained for
Extra refers to sub-experiments in the case of experiments 3-5.

Each folder contains another subfolder named as the ID of the training batch. Usually the date and time the training was started. The subfolder contains folders 1111, 2222, 3333, 4444, 5555. These are the random seeds used to initialise the 5 training runs per agent. Each of these subfolders then contain a "model" subfolder, which contains the pytorch weights (.pth) and hyperparameters (hyperparameters.json) of the trained agents. 

For example:
SAC_Auto_FiberCupNoWMTrainExp4:
                   -- 2023-03-22-07_38_30:
                                 --1111

                                 --2222

                                 --3333

                                 --4444

                                 --5555:
                                         --model:

                                                    --last_model_state_actor.pth

                                                    --last_model_state_critic.pth

                                                    --hyperparameters.json

Refer to https://github.com/scil-vital/TrackToLearn for usage.