"run_id": 0,
"env": 0,
"action_space_limits": ([-10.], [10.]),
"buffer_size": 10000,
"batch_size": 64,
"is_quanser_env": True,
"gamma": .99,
"tau": 1e-2,
"steps": 100000,
"warmup_samples": 1000,
"noise_decay": 0.9,
"transform": lambda x:x,
"actor_lr": 1e-3,
"critic_lr": 1e-3,
"lr_decay": 1.0,
"lr_min":1.e-7,
"trial_horizon": 5000,
"actor_hidden_layers":[10, 10, 10],
"critic_hidden_layers":[10, 10, 10],
"device": "cpu"
These will be set in every run, either a trial or training session!
run_id every algorithm should specify a run_id to obtain a structured order
env the environment which is used in this run (e.g. "Pendulum-v0")
action_space_limits limits the actions to select between two values, given in a tuple
buffer_size the buffer size taken of a replay buffer
batch_size size of batches that are sampled from a replay buffer, to train the actor and critic network with
is_quanser_env either if the used env is a quanser_robots environment (True) or not (False)
gamma used as a weight factor of the critic's prediction
tau soft update for the weights from the source to target network
steps that will be taken in this training session
warmup_samples number of samples generated before starting the training
noise_decay the gaussian noise will be reduced by this factor in every episode
transform user specific transformation function (for dimensionality reduction for example)
actor_lr adam optimizer learning rate of your actor networks
critic_lr adam optimizer learning rate of your critic networks
lr_decay decay of adam optimizers learning rates
lr_min your optimizer learning rate cant drop this value
trial_horizon in between training episodes number of trials to test your new policy
actor_hidden_layers a numeric list of layers between input and output layer that will be used in your actor networks
critic_hidden_layers a numeric list of layers between input and output layer that will be used in your critic networks
device either cpu or cuda (faster learning), tensors will be calculated on specified device