d3rlpy.algos.LinearDecayEpsilonGreedy

class d3rlpy.algos.LinearDecayEpsilonGreedy(start_epsilon=1.0, end_epsilon=0.1, duration=1000000)

\(\epsilon\)-greedy explorer with linear decay schedule.

Parameters

• start_epsilon (float) – Initial \(\epsilon\).

• end_epsilon (float) – Final \(\epsilon\).

• duration (int) – Scheduling duration.

Methods

compute_epsilon(step)

Returns decayed \(\epsilon\).

Returns

\(\epsilon\).

Parameters

step (int) –

Return type

float

sample(algo, x, step)

Returns \(\epsilon\)-greedy action.

Parameters

• algo (d3rlpy.interface.QLearningAlgoProtocol) – Algorithm.

• x (Union[numpy.ndarray[Any, numpy.dtype[Any]], Sequence[numpy.ndarray[Any, numpy.dtype[Any]]]]) – Observation.

• step (int) – Current environment step.

Returns

\(\epsilon\)-greedy action.

Return type

numpy.ndarray[Any, numpy.dtype[Any]]