torch_concepts.nn.InferenceOutput¶

class InferenceOutput(params: ~typing.Dict[str, ~typing.Dict[str, ~torch.Tensor]] = <factory>, guide_params: ~typing.Dict[str, ~typing.Dict[str, ~torch.Tensor]] = <factory>, samples: ~typing.Dict[str, ~torch.Tensor] = <factory>, probabilities: ~torch.Tensor | None = None)[source]¶

Return value of every inference engine.

params¶

Per-variable named parameter tensors of the model-side distribution (e.g. {'c': {'probs': ...}}).

Type:: dict[str, ParamDict]

guide_params¶

Per-latent named parameter tensors of the variational guide.

Type:: dict[str, ParamDict]

samples¶

Per-variable sampled values.

Type:: dict[str, torch.Tensor]

probabilities¶

Joint conditional probabilities for a fully realised query batch.

Type:: torch.Tensor or None

__init__(params: ~typing.Dict[str, ~typing.Dict[str, ~torch.Tensor]] = <factory>, guide_params: ~typing.Dict[str, ~typing.Dict[str, ~torch.Tensor]] = <factory>, samples: ~typing.Dict[str, ~torch.Tensor] = <factory>, probabilities: ~torch.Tensor | None = None) → None¶

Methods

__init__([params, guide_params, samples, ...])

Attributes

`probabilities`
`params`
`guide_params`
`samples`