Arguments for the Test part of the code Args: parser: the parser instance Returns: None: This function does not return a val

Returns all possible methods for predicting a y from samples of w.

Returns the names of available CLIP models

Returns the base bath where to log accuracies and tensorboard data. Returns: base_path (str): base path

Batchwise distance, adaptd from the original PCBM repository Args: self: instance samples1: first sample

Performs CBM inference Args: self: instance cs: concepts logits query (default=None): query Retu

Performs CBM inference Args: self: instance cs: concepts logits query (default=None): query Retu

Performs CBM inference Args: self: instance cs: concepts logits query (default=None): query Retu

CBM inference Args: self: instance pCs: concepts preds: predictions Returns: query_p

CBM inference Args: self: instance pCs: concepts preds: predictions Returns: query_p

CBM inference Args: self: instance pCs: concepts preds: predictions Returns: query_p

CBM inference Args: self: instance pCs: concepts preds: predictions Returns: query_p

Checks whether the dataset exists, if not creates it.

Checks whether the dataset exists, if not creates it.

Checks whether the dataset exists, if not creates it.

Performs inference inference Args: self: instance cs: concepts logits query (default=None): query

Performs inference inference Args: self: instance cs: concepts logits query (default=None): query

Performs inference inference Args: self: instance cs: concepts logits query (default=None): query

Performs inference inference Args: self: instance cs: concepts logits query (default=None): query

Combine queries Args: self: instance spreds: shapes predictions cpreds: colors predictions Retur

Combine queries Args: self: instance preds: predictions Returns: py: pattern probabilities

Compute distances between predicted embeddings and latents z Args: self: instance pred_embeddings: predicted embeddin

Computes query probability given the worlds probability P(w). Args: self: instance query: query worlds_pr

Computes query probability given the worlds probability P(w). Args: self: instance query: query worlds_pr

Computes query probability given the worlds probability P(w). Args: self: instance query: query worlds_pr

Computes query probability given the worlds probability P(w). Args: self: instance query: query worlds_pr

Computes query probability given the worlds probability P(w). Args: self: instance query: query worlds_pr

Computes query probability given the worlds probability P(w). Args: self: instance query: query worlds_pr

Computes query probability given the worlds probability P(w). Args: self: instance query: query worlds_pr

Conditional generation Args: model: network pc (default=None): probability of concepts Returns: out: decoder output

Convert applicable model parameters to fp16

Return the list of ADs necessary to describe an image with 'sequence_len' digits. 'n_facts' specifies how many digits we are considering (i.e

Build Worlds-Queries matrix

Build Worlds-Queries matrix

Build Worlds-Queries matrix

Build Worlds-Queries matrix

Build the ProbLog model using teh given facts, rules, evidence and query.

Traditional discrete diffusion sampler with fixed number of timesteps.

Extracts P(w) contained in the dictionary 'res' resulting from ProbLog model evaluation.

Parallel first-hitting sampler (See Zhang et al 2024) for NeSy masked diffusion.

:param concepts: different name of concepts :param layer_name: the name of the layer to compute CAV on :param activations: activations wi

Forward pass of DiT. This assumes w and y are one-hot encoded with the same size of final dimension You'll need to concatenat