Struct OffPolicyConfig 

pub struct OffPolicyConfig {
    pub num_envs: usize,
    pub autobatch_size: usize,
    pub replay_buffer_size: usize,
    pub train_interval: usize,
    pub train_steps: usize,
    pub eval_interval: usize,
    pub eval_episodes: usize,
    pub train_batch_size: usize,
    pub warmup_steps: usize,
}

Available on crate feature rl only.

Parameters of an on policy training with multi environments and double-batching.

Fields

num_envs: usize

The number of environments to run simultaneously for experience collection.

autobatch_size: usize

Number of environment state to accumulate before running one step of inference with the policy. Must be equal or less than the number of simultaneous environments.

replay_buffer_size: usize

Max number of transitions stored in the replay buffer.

train_interval: usize

The number of steps to collect between each step of training.

train_steps: usize

Number of optimization steps done each train_interval.

eval_interval: usize

The number of steps to collect between each evaluation.

eval_episodes: usize

The number of episodes to run for each evaluation.

train_batch_size: usize

The number of transition to train on.

warmup_steps: usize

Number of steps to collect before starting to train.

Implementations

impl OffPolicyConfig

pub fn new() -> Self

Create a new instance of the config.

Arguments

Default Arguments

num_envs

The number of environments to run simultaneously for experience collection.

• Defaults to 1

autobatch_size

Number of environment state to accumulate before running one step of inference with the policy. Must be equal or less than the number of simultaneous environments.

• Defaults to 1

replay_buffer_size

Max number of transitions stored in the replay buffer.

• Defaults to 1024

train_interval

The number of steps to collect between each step of training.

• Defaults to 1

train_steps

Number of optimization steps done each train_interval.

• Defaults to 1

eval_interval

The number of steps to collect between each evaluation.

• Defaults to 10_000

eval_episodes

The number of episodes to run for each evaluation.

• Defaults to 1

train_batch_size

The number of transition to train on.

• Defaults to 32

warmup_steps

Number of steps to collect before starting to train.

• Defaults to 0

impl OffPolicyConfig

pub fn with_num_envs(self, num_envs: usize) -> Self

Sets the value for the field num_envs.

The number of environments to run simultaneously for experience collection.

• Defaults to 1

pub fn with_autobatch_size(self, autobatch_size: usize) -> Self

Sets the value for the field autobatch_size.

Number of environment state to accumulate before running one step of inference with the policy. Must be equal or less than the number of simultaneous environments.

• Defaults to 1

pub fn with_replay_buffer_size(self, replay_buffer_size: usize) -> Self

Sets the value for the field replay_buffer_size.

Max number of transitions stored in the replay buffer.

• Defaults to 1024

pub fn with_train_interval(self, train_interval: usize) -> Self

Sets the value for the field train_interval.

The number of steps to collect between each step of training.

• Defaults to 1

pub fn with_train_steps(self, train_steps: usize) -> Self

Sets the value for the field train_steps.

Number of optimization steps done each train_interval.

• Defaults to 1

pub fn with_eval_interval(self, eval_interval: usize) -> Self

Sets the value for the field eval_interval.

The number of steps to collect between each evaluation.

• Defaults to 10_000

pub fn with_eval_episodes(self, eval_episodes: usize) -> Self

Sets the value for the field eval_episodes.

The number of episodes to run for each evaluation.

• Defaults to 1

pub fn with_train_batch_size(self, train_batch_size: usize) -> Self

Sets the value for the field train_batch_size.

The number of transition to train on.

• Defaults to 32

pub fn with_warmup_steps(self, warmup_steps: usize) -> Self

Sets the value for the field warmup_steps.

Number of steps to collect before starting to train.

• Defaults to 0

Trait Implementations

impl Clone for OffPolicyConfig

fn clone(&self) -> Self

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Config for OffPolicyConfig

fn save<P>(&self, file: P) -> Result<(), Error>

where P: AsRef<Path>,

Available on crate feature std only.

Saves the configuration to a file.

fn load<P>(file: P) -> Result<Self, ConfigError>

where P: AsRef<Path>,

Available on crate feature std only.

Loads the configuration from a file.

fn load_binary(data: &[u8]) -> Result<Self, ConfigError>

Loads the configuration from a binary buffer.

impl Debug for OffPolicyConfig

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for OffPolicyConfig

fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>

where D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Display for OffPolicyConfig

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Serialize for OffPolicyConfig

fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>

where S: Serializer,

Serialize this value into the given Serde serializer.