H5matref: obtain an HDF5 dataset reference suitable for handling as...

In vjcitn/BiocSklearn: interface to python sklearn via Rstudio reticulate

obtain an HDF5 dataset reference suitable for handling as numpy matrix

Description

obtain an HDF5 dataset reference suitable for handling as numpy matrix

Usage

H5matref(filename, dsname = "assay001")

Arguments

filename	a pathname to an HDF5 file
dsname	internal name of HDF5 matrix to use, defaults to 'assay001'

Value

instance of (S3) "h5py._hl.dataset.Dataset"

Note

This should only be used with persistent environment discipline of basilisk. Additional support is planned in Bioc 3.12.

Examples

fn = system.file("ban_6_17/assays.h5", package="BiocSklearn")
ban = H5matref(fn)
ban
proc = basilisk::basiliskStart(BiocSklearn:::bsklenv)
fullpca = basilisk::basiliskRun(proc, function() {
 np = import("numpy", convert=FALSE) # ensure
 print(ban$shape)
 print(np$take(ban, 0:3, 0L))
 fullpca = skPCA(ban)
 tx = getTransformed(fullpca)
 print(dim(tx))
 fullpca
 })
basilisk::basiliskStop(proc)
# project samples
np = reticulate::import("numpy", convert=FALSE, delay_load=TRUE)
np$take(ban, 0:20, 0L)$shape
st = skPartialPCA_step(np$take(ban, 0:20, 0L), n_comp=4L)
st = skPartialPCA_step(np$take(ban, 21:40, 0L), n_comp=4L, obj=st)
st = skPartialPCA_step(np$take(ban, 41:63, 0L), n_comp=4L, obj=st)
oo = st$transform(ban)
dim(oo)
cor(oo[,1:4], getTransformed(fullpca)[,1:4])

vjcitn/BiocSklearn documentation built on Aug. 30, 2024, 2:13 p.m.