decompose_tensor: Wrapper function to run 'cell2cell_tensor' decomposition on a...

In saezlab/ligrec_decoupler: LIANA: a LIgand-receptor ANalysis frAmework

Wrapper function to run 'cell2cell_tensor' decomposition on a prebuilt tensor.

Description

Wrapper function to run 'cell2cell_tensor' decomposition on a prebuilt tensor.

Usage

decompose_tensor(
  tensor,
  rank = NULL,
  tf_optimization = "robust",
  seed = 1337,
  upper_rank = 25,
  elbow_metric = "error",
  smooth_elbow = FALSE,
  init = "svd",
  svd = "numpy_svd",
  factors_only = TRUE,
  verbose = TRUE,
  ...
)

Arguments

tensor	Tensor-cell2cell Prebuilt.Tensor class instance
rank	Ranks for the Tensor Factorization (number of factors to deconvolve the original tensor). If NULL, then rank selection is performed using the 'elbow_rank_selection' function.
tf_optimization	indicates whether running the analysis in the ''regular'‘ or the '’robust'' way. The regular way means that the tensor decomposition is run 10 times per rank evaluated in the elbow analysis, and 1 time in the final decomposition. Additionally, the optimization algorithm has less number of iterations in the regular than the robust case (100 vs 500) and less precision (tolerance of 1e-7 vs 1e-8). The robust case runs the tensor decomposition 20 times per rank evaluated in the elbow analysis, and 100 times in the final decomposition. Here we could use the ‘tf_optimization=’regular'', which is faster but generates less robust results. We recommend using ‘tf_optimization=’robust', which takes longer to run (more iteractions and more precise too).
seed	Random seed integer
upper_rank	Upper bound of ranks to explore with the elbow analysis.
init	Initialization method for computing the Tensor Factorization. ‘svd’, ‘random’
factors_only	whether to return only the factors after factorization
verbose	verbosity logical
...	Dictionary containing keyword arguments for the c2c.compute_tensor_factorization function. The function deals with 'random_state' (seed) and 'rank' internally.

Value

an instance of the cell2cell.tensor.BaseTensor class (via reticulate). If build_only is TRUE, then no rank selection or tensor decomposition is returned. Otherwise, returns a tensor with factorization results.

saezlab/ligrec_decoupler documentation built on Aug. 10, 2024, 8:56 a.m.