grid_size: int of the grid height and width return: pos_embed: [grid_size*grid_size, embed_dim] or [1+grid_size*grid_size, embed_dim] (w/

The function to generate the activations of all layers for ONE concept only :param model: :param output_dir: :param data_loader: the

Returns the output probability of the model with the classifier

Main function. Provides functionalities for training, testing and active learning. Args: args: parsed command line arguments. Return

Mean Entropy Args: probabilities (ndarray): probability vector n_values (int): n values Returns: entropy_values: mea

Computes the probability for each ProbLog fact given the latent vector z

Computes the probability for each ProbLog fact given the latent vector z Args: self: instance z (torch.tensor): laten

Computes the probability for each ProbLog fact given the latent vector z Args: self: instance z (torch.tensor): laten

Computes the probability for each ProbLog fact given the latent vector z Args: self: instance z (torch.tensor): laten

Computes the probability for each ProbLog fact given the latent vector z Args: self: instance z (torch.tensor): laten

Computes the probability for each ProbLog fact given the latent vector z

Computes the probability for each ProbLog fact given the latent vector z Args: self: instance concepts (torch.tensor)

Computes the probability for each ProbLog fact given the latent vector z Args: self: instance z (torch.tensor): laten

Computes the probability for each ProbLog fact given the latent vector z Args: self: instance z (torch.tensor): laten

Computes the probability for each ProbLog fact given the latent vector z Args: self: instance concepts (torch.tensor)

Computes the probability for each ProbLog fact given the latent vector z Args: self: instance z (torch.tensor): laten

Computes the probability for each ProbLog fact given the latent vector z Args: self: instance concepts (torch.tensor)

Computes the probability for each ProbLog fact given the latent vector z Args: self: instance, z: latents

Computes the probability for each ProbLog fact given the latent vector z Args: self: instance z: latents

Computes the probability for each ProbLog fact given the latent vector z Args: self: instance z (toch.tensor): latent

Computes the probability for each ProbLog fact given the latent vector z Args: self: instance z (torch.tensor): laten

Computes the probability for each ProbLog fact given the latent vector z Args: self: instance z (torch.tensor): laten

Computes the probability for each ProbLog fact given the latent vector z Args: self: instance z (torch.tensor): laten

Computes the probability for each ProbLog fact given the latent vector z Args: self: instance concepts (torch.tensor)

Computes the probability for each ProbLog fact given the latent vector z Args: self: instance z (torch.tensor): laten

Computes the probability for each ProbLog fact given the latent vector z

Computes the probability for each ProbLog fact given the latent vector z

Returns the identity element of the multiplication.

Parse command line arguments Returns: args: parsed command line arguments

Preprocess Kandinksy images Args: model: network dataset: dataset args: command line arguments Returns: None:

TRAINING Args: model (MnistDPL): network dataset (BaseDataset): dataset _loss (ADDMNIST_DPL): loss function args:

Performs ProbLog inference to retrieve the worlds probability distribution P(w). Works with two encoded bits. Args: self: instanc

Performs ProbLog inference to retrieve the worlds probability distribution P(w). Works with an arbitrary number of encoded bits (digits).

Problog inference Args: self: instance pCs: probability of concepts query (default=None): query

Performs ProbLog inference to retrieve the worlds probability distribution P(w). Works with two encoded bits. Args: self: instanc

Performs ProbLog inference to retrieve the worlds probability distribution P(w). Works with two encoded bits. Args: self: instanc

Problog inference Args: self: instance pCs: probabilities of concepts preds: predictions Returns

Performs ProbLog inference to retrieve the worlds probability distribution P(w) and the desired query probability.

Returns images and labels

Returns images and labels

Returns images and labels

Recon visualization method Args: out_dict: output dictionary Returns: out: images and recons concatenated

Conditional reinforce with leave-one-out (rloo) loss. Always adds w^0 to the set of samples, renormalising. log_probs: predicted log

Reinforce with leave-one-out (rloo) loss. log_probs: predicted log probabilities for each action taken rewards: resulting re