Struct LinearSoftmaxPolicy 

pub struct LinearSoftmaxPolicy<T: Float + Debug + Send + Sync + 'static> { /* private fields */ }

Categorical policy π(a|s) = softmax(W φ(s)) with an analytic score function.

Actions are one-hot rows of width n_actions. The score function is the textbook softmax result

∇_W log π(a|s) = (e_a − p(s)) φ(s)ᵀ ,  p(s) = softmax(W φ(s))

which is exactly rank one — so this policy is also an exact source of Kronecker factors for K-FAC (φ is the input factor, e_a − p the output factor).

Implementations

impl<T: Float + Debug + Send + Sync + 'static> LinearSoftmaxPolicy<T>

pub fn new(n_actions: usize, n_features: usize) -> Result<Self>

Create a zero-initialized policy (uniform over actions).

pub fn from_weights(weights: Array2<T>) -> Result<Self>

Create a policy from an explicit weight matrix (n_actions, n_features).

pub fn n_actions(&self) -> usize

Number of discrete actions.

pub fn n_features(&self) -> usize

Observation/feature dimension.

pub fn weights(&self) -> &Array2<T>

Current logit weights.

pub fn logits(&self, observations: &Array2<T>) -> Result<Array2<T>>

Logits W φ(s) for a batch, shape (n, n_actions).

pub fn probabilities(&self, observations: &Array2<T>) -> Result<Array2<T>>

Action probabilities for a batch, shape (n, n_actions).

pub fn one_hot(&self, index: usize) -> Result<Array1<T>>

One-hot encode an action index.

pub fn sample_actions_with( &self, observations: &Array2<T>, uniform: impl FnMut() -> f64, ) -> Result<(Array2<T>, Array1<T>, Vec<usize>)>

Sample one action per observation using a caller-supplied uniform source.

Returns the one-hot action matrix together with the log-probabilities of the sampled actions — exactly the two arrays a super::TrajectoryBatch needs. Taking the uniform source as a closure keeps rollouts fully reproducible in tests.

Trait Implementations

impl<T: Clone + Float + Debug + Send + Sync + 'static> Clone for LinearSoftmaxPolicy<T>

fn clone(&self) -> LinearSoftmaxPolicy<T>

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl<T: Debug + Float + Debug + Send + Sync + 'static> Debug for LinearSoftmaxPolicy<T>

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

Formats the value using the given formatter.

impl<T: Float + Debug + Send + Sync + 'static> PolicyNetwork<T> for LinearSoftmaxPolicy<T>

fn evaluate_actions( &self, observations: &Array2<T>, actions: &Array2<T>, ) -> Result<PolicyEvaluation<T>>

Evaluate actions for given observations

fn get_action_distribution( &self, observations: &Array2<T>, ) -> Result<ActionDistribution<T>>

Get action distribution for given observations

fn update_parameters( &mut self, deltas: &HashMap<String, Array1<T>>, ) -> Result<()>

Add a parameter delta to the policy parameters (see the trait docs).

fn get_parameters(&self) -> HashMap<String, Array1<T>>

Get current policy parameters

fn log_prob_gradient( &self, observations: &Array2<T>, actions: &Array2<T>, coefficients: &Array1<T>, ) -> Result<HashMap<String, Array1<T>>>

Gradient of a coefficient-weighted sum of log-probabilities: ∂/∂θ Σᵢ cᵢ · log π(aᵢ | sᵢ).

fn entropy_gradient( &self, observations: &Array2<T>, ) -> Result<HashMap<String, Array1<T>>>

Gradient of the batch-mean entropy ∂/∂θ (1/N) Σᵢ H[π(·|sᵢ)].

fn score_matrix( &self, observations: &Array2<T>, actions: &Array2<T>, ) -> Result<Array2<T>>

Per-sample score vectors g_i = ∇_θ log π(aᵢ|sᵢ), one per row, flattened with flatten_named’s canonical layout.

fn kronecker_factors( &self, observations: &Array2<T>, actions: &Array2<T>, ) -> Result<Vec<KroneckerBlock<T>>>

Per-sample Kronecker factors of the Fisher information matrix.

fn mean_action_gradient( &self, observations: &Array2<T>, weights: &Array2<T>, ) -> Result<HashMap<String, Array1<T>>>

Gradient of a weighted sum of the distribution mean: ∂/∂θ Σᵢ Σⱼ w[i,j] · μⱼ(sᵢ).