Obtain some examples of your algorithms.
The algorithm is explained right here -> DDPG.
To start a trial session you can for example use this command.
python main.py ddpg sim trial src/config/example/trial src/config/example/trial_out 1
The command will save a reward plot in this directory in the folder out as you may test by yourself.
Your specified folder needs always to specify a parameters.csv and a policy (the name is mandatory).
Where policy defines a model (neural network weights) for the policy and parameters are the used parameters
to load the model and execute either in training or in trial mode.
This example was executed in trial mode as you can read in the command.
To start a training session you can for example use this command.
python main.py ddpg sim train src/config/example/train/parameters.csv src/config/example/train_out
After that you can obtain a actortarget model that fits your needs and take the parameters.csv to a new folder
and execute it as in the trial section.
The algorithm is explained right here -> MPC
To start a training session you can use this command.
python main.py mpc sim train src/config/example/train/testmpc.csv src/config/example/train_out
After that src/config/example/train_out/textmpc_CartpoleStabShort-v0 will contain the results of the training. This is a file rewarddata containing the total episode rewards and a file trajectories containing all trial trajectories, each on one line, in order.