autoFilterCV: Cross-validation filtering of autoencoder

In jingshuw/SAVERX: Transfer Learning for Single-cell RNA-seq Denoising

Cross-validation filtering of autoencoder

Description

Cross-validation is done to determine which genes can not be predicted well, by comparing the autoencoder predicted loss with the loss estimating the gene expression as a constant across cells

Usage

autoFilterCV(x, python.module, main, pretrain_file = "",
  nonmissing_indicator = 1, n_human = 21183L, n_mouse = 21122L,
  shared_size = 15494L, model.species = NULL, out_dir = ".",
  batch_size = 32L, write_output_to_tsv = F, fold = 6, samp = 3,
  epsilon = 1e-10, seed = 1, ...)

Arguments

x	Target sparse data matrix of gene by cell. When pretraining is used, the genes should be the same as the nodes used in the pretrained model. If a node gene is missing is the target dataset, set all values of that gene as 0 in x and indicate that using nonmissing_indicator
python.module	The python module for the Python package sctransfer
main	A Python main module
pretrain_file	The pretrained weights file ended with '.hdf5'
nonmissing_indicator	A single value 1 or a vector of 0 and 1s to indicate which nodes are missing in the target dataset. Set to 1 for no pretraining.
model.species	Should be either 'Human' or 'Mouse' when pretraining is used
write_output_to_tsv	If True, then the result of Python is written as .tsv files instead of passing back to R. Default is False.
fold	Number of total CV folds
samp	Number of sampled folds taken to reduce CV time cost
...	Extra parameters passed to Python module sctransfer function api (if no pretraining) or function api_pretrain (with pretraining).

Value

a list of the filtered predicted data matrix and the CV error

jingshuw/SAVERX documentation built on June 20, 2022, 3:17 a.m.