sakura.utils.distributions.rand_uniform

sakura.utils.distributions.rand_uniform(batch_size, n_dim=2, low=-1.0, high=1.0, n_labels=1, label_offsets=None, label_indices=None) → Tensor

Generates samples from a uniform distribution with optional label-based offsets.

When n_labels > 1, applies label-specific offsets to create separated clusters.

Parameters:

• batch_size (int) – Number of batch samples

• n_dim (int, optional) – Number of latent dimension, defaults to 2

• low (float, optional) – Lower bound of uniform distribution (for each dimension), defaults to -1.

• high (float, optional) – Upper bound of uniform distribution (for each dimension), defaults to 1.

• n_labels (int, optional) – Number of labels to consider in supervision, defaults to 1, where supervision is off

• label_offsets (list[list[float]], optional) – Offsets for each label cluster

• label_indices (list[int], optional) – List of label indices (0 <= index < n_labels) for each batch sample, randomly assigned if none provided

Returns:

Tensor of shape (batch_size, n_dim) with uniform samples

Return type:

torch.FloatTensor