mira.topics.gradient_tune#
- mira.topics.gradient_tune(model, data, max_attempts=3, max_topics=None)#
Tune number of topcis using a gradient-based estimator based on the Dirichlet Process model. This tuner is very fast, though less comprehensive than the BayesianTuner. We recommend using this tuner for large datasets (>40K cells).
- Parameters
- modelmira.topics.TopicModel
Topic model to tune. The provided model should have columns specified to retrieve endogenous and exogenous features, and should have the learning rate configued by
get_learning_rate_bounds.- dataanndata.AnnData
Anndata of expression or accessibility data.
- max_topicsint > 0 or None
If none, MIRA automatically chooses an upper limit on the number of topics to model based on a generous hueristic calculated from the number of cells in the provided dataset. If a value is provided, that upper limit is used instead.
- Returns
- num_topicsint
Estimated number of topics in dataset
- max_topic_contributionsnp.ndarray[float] of shape (n_topics,)
For each topic attempted to learn from the data, its maximum contribution to any cell.