TruncExp#

class braintools.cogtask.TruncExp(mean, min_val=0, max_val=inf, key=None)[source]#

Truncated exponential distribution for sampling time durations.

Useful for generating variable-duration phases with exponential distribution truncated to a specified range.

Examples

>>> import brainunit as u
>>> import jax
>>> import numpy as np
>>> from braintools.cogtask import (
...     Feature, Fixation, Response, Task, TruncExp, VariableDuration)
>>> delay = TruncExp(600 * u.ms, 300 * u.ms, 1500 * u.ms, key=jax.random.PRNGKey(0))
>>> delay
TruncExp(mean=600ms, min=300ms, max=1500ms)
>>> delay.min_value(), delay.max_value()
(Quantity(300, "ms"), Quantity(1500, "ms"))
>>> d = delay()  # one sample from the sampler's own key
>>> bool(300 * u.ms <= d) and bool(d < 1500 * u.ms)
True
>>>
>>> # Typical use: sample in ``trial_init`` with the trial's RNG and
>>> # read the value in a ``VariableDuration`` phase.
>>> task = Task(
...     Fixation(100 * u.ms, inputs={'fixation': 1.0})
...     >> VariableDuration(300 * u.ms, 1500 * u.ms, ctx_key='delay',
...                         inputs={'fixation': 1.0})
...     >> Response(100 * u.ms, outputs={'label': 1}),
...     Feature(1, 'fixation'), Feature(2, 'choice'),
...     trial_init=lambda ctx: ctx.update(delay=delay(ctx)),
...     dt=10.0 * u.ms, seed=0,
... )
>>> task.max_trial_duration()  # 10 + 150 + 10 steps
170
>>> X, Y, info = task.sample_trial(0)
>>> X.shape, 30 <= int(np.asarray(info['mask']).sum()) <= 170
((170, 1), True)
max_value()[source]#

Truncation upper bound in the sampler’s time unit.

min_value()[source]#

Truncation lower bound in the sampler’s time unit.