Public
Edited
Apr 15, 2019
Importers
4 stars
chooseAction = (env, observation, policyFn) => {
const actionProbabilities = policyFn(env, observation);
return env.actions[chooseRandomly(actionProbabilities)];
}
function* genEpisodeStep(env, policyFn, episodeNumber, limit) {
let { state: state, observation: observation } = reset(env);
let action = chooseAction(env, observation, policyFn);
let done = false,
i = 0;
while (!done && i < limit) {
const result = step(env, state, action);
const nextAction = !result.done && chooseAction(env, result.observation, policyFn);
yield {
current: {
observation: observation,
action: action,
_state: state
// Technically, the agent can not directly access the current state;
// it can only make observations that are possibly noisy measurements of
// the current state -- but it's convenient to keep it here for
// visualization purposes, so we slap the underscore in front here.
},
reward: result.reward,
next: {
observation: result.observation,
action: nextAction,
_state: result.state // technically, state is hidden... hence the underscore
},
done: result.done,
episodeNumber: episodeNumber,
stepNumber: i
};
done = result.done;
state = result.state;
observation = result.observation;
action = nextAction;
++i;
}
}
// episode = (policy, limit) => [...genEpisodeStep(env, policy, limit)]
function* genEpisode(env, policy, limit) {
let epi = [];
for (const aStep of genEpisodeStep(env, policy, epi.length, limit)) {
epi.push(aStep);
yield epi;
}
}
function* slowGenEpisode(env, policyFn, limit=100, interval=1000) {
yield* slowly(() => genEpisode(env, policyFn, limit), interval);
// for (const aStep of genEpisode(env, policyFn, limit))
// yield Promises.tick(interval, aStep);
}
Purpose-built for displays of data
Observable is your go-to platform for exploring data and creating expressive data visualizations. Use reactive JavaScript notebooks for prototyping and a collaborative canvas for visual data exploration and dashboard creation.
