Skip to content

Categorical¤

distreqx.distributions.Categorical(distreqx.distributions.AbstractSTDDistribution, distreqx.distributions.AbstractSampleLogProbDistribution, distreqx.distributions.AbstractSurvivalDistribution) ¤

Categorical distribution over integers.

The Categorical distribution is parameterized by either probabilities (probs) or unormalized log-probabilities (logits) of a set of \(K\) classes. It is defined over the integers \(\{0, 1, \ldots, K - 1\}\).

__init__(logits: Array | None = None, probs: Array | None = None) ¤

Initializes a Categorical distribution.

Arguments:

  • logits: Logit transform of the probability of each category. Only one of logits or probs can be specified.
  • probs: Probability of each category. Only one of logits or probs can be specified.

distreqx.distributions.OneHotCategorical(distreqx.distributions.AbstractSTDDistribution, distreqx.distributions.AbstractSampleLogProbDistribution, distreqx.distributions.AbstractSurvivalDistribution) ¤

OneHotCategorical distribution.

__init__(logits: Array | None = None, probs: Array | None = None) ¤

Initializes a OneHotCategorical distribution.

Arguments:

  • logits: Logit transform of the probability of each category. Only one of logits or probs can be specified.
  • probs: Probability of each category. Only one of logits or probs can be specified.