StationaryOTKernel.compute_transition_matrix(eps, dt, basis='X_pca', cost_norm_method=None, method='ent', tol=0.0, thresh=0.0, maxiter=5000, C=None, verbose=False, **kwargs)[source]

Compute transition matrix using stationary OT [Zhang et al., 2021].

  • eps (float) – Regularization parameter.

  • dt (float) – Choice of the time step over which to fit the model.

  • basis (str) – Key in anndata.AnnData.obsm where the basis is stored.

  • cost_norm_method (Optional[str]) – Cost normalization method to use. Use “mean” to ensure mean(C) = 1 or refer to ot.utils.cost_normalization() for more information.

  • method (Literal[‘ent’, ‘quad’, ‘unbal’]) –

    Choice of regularization. Valid options are:

    • ’ent’ - entropy.

    • ’quad’ - L2-norm.

    • ’unbal’ - unbalanced transport (not yet implemented).

  • tol (float) – Relative tolerance for OT solver convergence.

  • thresh (float) – Threshold for output transition probabilities.

  • maxiter (int) – Maximum number of iterations for OT solver.

  • C (Optional[ndarray]) – Cost matrix for optimal transport problem.

  • verbose (bool) – Detailed output on convergence of OT solver.

  • kwargs (Any) – Additional keyword arguments.

Return type



cellrank.external.kernels.StationaryOTKernel Makes transition_matrix available.