rlmesh.trial_index

functionfrom rlmesh import trial_index[source]

The reserved trial_index reset option, or None when absent.

trial_index(
    options: Mapping[str, Any] | None,
    *,
    lane: int | None = None,
) -> int | None

The ordinal of the episode a reset starts, 0-based and walked in order by the runtime (from run(trial_index_base=...), 0 by default), so an env can sweep a fixed list of initial states / goals exactly as its upstream benchmark does instead of re-deriving one from a hashed seed. Delivered only to an env that declared "trial_index" in EnvFactory.reset_options; absent on a hand-driven Session.reset() that passed none::

class MyEnv(rlmesh.EnvFactory): reset_options = (“trial_index”,)

def reset(self, *, seed=None, options=None): trial = rlmesh.trial_index(options) n = len(self.init_states) state = self.init_states[(trial if trial is not None else seed or 0) % n]

The option takes one of two forms. A reset covering one lane carries a bare integer; a reset covering several lanes of a vector env carries a list, one ordinal per lane being reset, in the same order as the reset’s seeds. A vector env reads its lanes with lane=, the position among the lanes this reset covers (not the env index, when only some lanes are reset)::

for i in range(len(lanes_being_reset)): trial = rlmesh.trial_index(options, lane=i)

Parameters

Name	Type	Description
`options`	`Mapping[str, Any] \| None`	The `options` mapping `reset` received, or `None`.
`lane`	`int \| None`	Position among the lanes this reset covers. `None` (the default) reads the single-lane form; a bare integer also answers `lane=0`.

Returns

Type	Description
`int \| None`	The ordinal, or `None` when the key is absent or not an integer.

Raises

Type	Description
`ValueError`	The option is the multi-lane list and `lane` was not given, or `lane` is outside the lanes the option covers.