@ai-on-browser/data-analysis-models
Version:
Data analysis model package without any dependencies
65 lines (64 loc) • 2.02 kB
TypeScript
/**
* A2C agent
*/
export default class A2CAgent {
/**
* @param {RLEnvironmentBase} env Environment
* @param {number} resolution Resolution of actions
* @param {number} procs Number of processes
* @param {LayerObject[]} layers Network layers
* @param {string} optimizer Optimizer of the network
*/
constructor(env: RLEnvironmentBase, resolution: number, procs: number, layers: LayerObject[], optimizer: string);
_net: ActorCriticNet;
_procs: number;
_env: RLEnvironmentBase;
_advanced_step: number;
_gamma: number;
_states: void[];
_envs: RLEnvironmentBase[];
terminate(): void;
/**
* Returns a score.
* @returns {Array<Array<Array<number>>>} Score values
*/
get_score(): Array<Array<Array<number>>>;
/**
* Returns a action.
* @param {*[]} state Current states
* @returns {*[]} Action
*/
get_action(state: any[]): any[];
/**
* Update model.
* @param {boolean} done Done epoch or not
* @param {number} learning_rate Learning rate
* @param {number} batch Batch size
* @returns {number} Loss value
*/
update(done: boolean, learning_rate: number, batch: number): number;
}
export type LayerObject = import("./nns/graph").LayerObject;
/**
* @ignore
* @typedef {import("./nns/graph").LayerObject} LayerObject
*/
declare class ActorCriticNet {
constructor(env: any, resolution?: number, layers?: any[], optimizer?: string);
_resolution: number;
_states: any;
_actions: any;
_action_sizes: any;
_layers: any[];
_net: NeuralNetwork;
get_action(state: any): any[];
_state_to_input(s: any): any[];
get_score(state: any): any;
_states_data: any[];
_action_pos(action: any): number;
_pos_action(i: any): any[];
update(states: any, actions: any, rewards: any, learning_rate: any, batch: any): number;
}
import { RLEnvironmentBase } from '../rl/base.js';
import NeuralNetwork from './neuralnetwork.js';
export {};