Continuous Reinforcement Learning

Specialization of the Reinforcement Learning Problem for continuous action domains.

Usage

e["Problem"]["Type"] = "ReinforcementLearning/Continuous"

Compatible Solvers

This problem can be solved using the following modules:

Variable-Specific Settings

These are settings required by this module that are added to each of the experiment’s variables when this module is selected.

Type

  • Usage: e[“Variables”][index][“Type”] = string

  • Description: Indicates if the variable belongs to the state or action vector.

  • Options:

    • “State”: The variable describes a state.

    • “Action”: The variable describes an action.

Lower Bound

  • Usage: e[“Variables”][index][“Lower Bound”] = float

  • Description: Lower bound for the variable’s value.

Upper Bound

  • Usage: e[“Variables”][index][“Upper Bound”] = float

  • Description: Upper bound for the variable’s value.

Name

  • Usage: e[“Variables”][index][“Name”] = string

  • Description: Defines the name of the variable.

Configuration

These are settings required by this module.

Agents Per Environment

  • Usage: e[“Problem”][“Agents Per Environment”] = unsigned integer

  • Description: Number of agents in a given environment. All agents share the same policy .

Environment Function

  • Usage: e[“Problem”][“Environment Function”] = Computational Model

  • Description: Function to initialize and run an episode in the environment.

Actions Between Policy Updates

  • Usage: e[“Problem”][“Actions Between Policy Updates”] = unsigned integer

  • Description: Number of actions to take before requesting a new policy.

Testing Frequency

  • Usage: e[“Problem”][“Testing Frequency”] = unsigned integer

  • Description: Number of generations after which the policy will be forcibly tested (even if it does not meet the threshold).

Training Reward Threshold

  • Usage: e[“Problem”][“Training Reward Threshold”] = float

  • Description: Minimum value of the episode’s cummulative sum of rewards for a policy to be considered as candidate.

Policy Testing Episodes

  • Usage: e[“Problem”][“Policy Testing Episodes”] = unsigned integer

  • Description: Number of test episodes to run the policy (without noise) for, for which the average average sum of rewards will serve to evaluate the termination criteria.

Custom Settings

  • Usage: e[“Problem”][“Custom Settings”] = knlohmann::json

  • Description: Any used-defined settings required by the environment.

Default Configuration

These following configuration will be assigned by default. Any settings defined by the user will override the given settings specified in these defaults.

{
"Actions Between Policy Updates": 0,
"Agents Per Environment": 1,
"Custom Settings": {    },
"Policy Testing Episodes": 5,
"Testing Frequency": 0
}

Variable Defaults

These following configuration will be assigned to each of the experiment variables by default. Any settings defined by the user will override the given settings specified in these defaults.

{
"Lower Bound": -Infinity,
"Type": "State",
"Upper Bound": Infinity
}