class pymc.Truncated(name, *args, rng=None, dims=None, initval=None, observed=None, total_size=None, transform=UNSET, **kwargs)[source]#

Truncated distribution

The pdf of a Truncated distribution is

\[\begin{split}\begin{cases} 0 & \text{for } x < lower, \\ \frac{\text{PDF}(x, dist)}{\text{CDF}(upper, dist) - \text{CDF}(lower, dist)} & \text{for } lower <= x <= upper, \\ 0 & \text{for } x > upper, \end{cases}\end{split}\]
dist: unnamed distribution

Univariate distribution created via the .dist() API, which will be truncated. This distribution must be a pure RandomVariable and have a logcdf method implemented for MCMC sampling.


dist will be cloned, rendering it independent of the one passed as input.

lower: tensor_like of float or None

Lower (left) truncation point. If None the distribution will not be left truncated.

upper: tensor_like of float or None

Upper (right) truncation point. If None, the distribution will not be right truncated.

max_n_steps: int, defaults 10_000

Maximum number of resamples that are attempted when performing rejection sampling. A TruncationError is raised if convergence is not reached after that many steps.

truncated_distribution: TensorVariable

Graph representing a truncated RandomVariable. A specialized Op may be used if the Op of the dist has a dispatched _truncated function. Otherwise, a SymbolicRandomVariable graph representing the truncation process, via inverse CDF sampling (if the underlying dist has a logcdf method), or rejection sampling is returned.


with pm.Model():
    normal_dist = pm.Normal.dist(mu=0.0, sigma=1.0)
    truncated_normal = pm.Truncated("truncated_normal", normal_dist, lower=-1, upper=1)


Truncated.dist(dist[, lower, upper, max_n_steps])

Creates a tensor variable corresponding to the cls distribution.