package fehu

You can search for identifiers within the package.

in-package search v0.2.0

Reinforcement learning framework for OCaml

Install

dune-project

Dependency

github.com Readme Changelog Edit opam file Versions (2)

Authors

Thibaut Mattio

Maintainers

Thibaut Mattio

Sources

raven-1.0.0.alpha2.tbz

sha256=93abc49d075a1754442ccf495645bc4fdc83e4c66391ec8aca8fa15d2b4f44d2

sha512=5eb958c51f30ae46abded4c96f48d1825f79c7ce03f975f9a6237cdfed0d62c0b4a0774296694def391573d849d1f869919c49008acffca95946b818ad325f6f

doc/fehu.visualize/Fehu_visualize/index.html

Module `Fehu_visualize`Source

Visualization sinks, encoders, and rollout helpers.

Sourcemodule Overlay : sig ... end

Overlays for augmenting rendered frames with diagnostic annotations.

Sourcemodule Video : sig ... end

Video recording wrappers for environments and vectorized environments.

Sourcemodule Sink : sig ... end

Sourceval push : Sink.t -> Fehu.Render.t -> unit

Push a single frame to the sink.

Sourceval push_many : Sink.t -> Fehu.Render.t array -> unit

Push multiple frames sequentially.

Source

val record_rollout : 
  env:('obs, 'act, Fehu.Render.t) Fehu.Env.t ->
  policy:('obs -> 'act * float option * float option) ->
  steps:int ->
  ?overlay:Overlay.t ->
  sink:Sink.t ->
  unit ->
  unit

Record a rollout from a single environment to the sink.

Source

val record_evaluation : 
  vec_env:('obs, 'act, Fehu.Render.t) Fehu.Vector_env.t ->
  policy:('obs array -> 'act array * float array option * float array option) ->
  n_episodes:int ->
  ?max_steps:int ->
  layout:[ `Single_each | `NxM_grid of int * int ] ->
  ?overlay:Overlay.t ->
  sink:Sink.t ->
  unit ->
  Fehu.Training.stats

Evaluate a vectorized policy while recording frames.