Utilities

tinytopics.utils

set_random_seed(seed)

Set the random seed for reproducibility across Torch and NumPy.

Parameters:

Name	Type	Description	Default
seed	int	Random seed value.	required

generate_synthetic_data(n, m, k, avg_doc_length=1000, device=None)

Generate synthetic document-term matrix for testing the model.

Parameters:

Name	Type	Description	Default
n	int	Number of documents.	required
m	int	Number of terms (vocabulary size).	required
k	int	Number of topics.	required
avg_doc_length	int	Average number of terms per document. Default is 1000.	1000
device	device	Device to place the output tensors on.	None

Returns:

Type	Description
Tensor	Document-term matrix.
ndarray	True document-topic distribution (L).
ndarray	True topic-term distribution (F).

align_topics(true_F, learned_F)

Align learned topics with true topics for visualization, using cosine similarity and linear sum assignment.

Parameters:

Name	Type	Description	Default
true_F	ndarray	Ground truth topic-term matrix.	required
learned_F	ndarray	Learned topic-term matrix.	required

Returns:

Type	Description
ndarray	Permutation of learned topics aligned with true topics.

sort_documents(L_matrix)

Sort documents grouped by dominant topics for visualization.

Parameters:

Name	Type	Description	Default
L_matrix	ndarray	Document-topic distribution matrix.	required

Returns:

Type	Description
Sequence[int]	Sequence[int]: Indices of documents sorted by dominant topics.