pas ssparse matrices between R and Python

serialization.py

"""
input and output some tricky matrix data in a cross-language format
"""

from scipy.sparse import coo_matrix, dok_matrix, csc_matrix
import tables
import numpy as np

def write_sparse_hdf(handle, group, data, colnames=None, filt=None):
    """
    sparse CSC (compressed sparse colums) matrices via hdf5
    32 bit for now, ok?
    """
    data_atom_type = tables.Float32Atom()
    if np.issubdtype(data.dtype, int):
        data_atom_type = tables.Int32Atom()
    handle.create_carray(group,'v_indices',
        atom=tables.Int32Atom(), shape=data.indices.shape,
        title="indices",
        filters=filt)[:] = data.indices
    handle.create_carray(group,'v_indptr',
        atom=tables.Int32Atom(), shape=data.indptr.shape,
        title="index ptr",
        filters=filt)[:] = data.indptr
    handle.create_carray(group,'v_data',
        atom=data_atom_type, shape=data.data.shape,
        title="data",
        filters=filt)[:] = data.data
    handle.create_carray(group,'v_datadims',
        atom=tables.Int32Atom(), shape=(2,),
        title="data dims",
        filters=filt)[:] = data.shape
    if colnames:
        handle.create_carray(group,'v_col_names',
            atom=tables.StringAtom(
                max([len(n) for n in colnames])
            ), shape=(len(colnames),),
            title="col names",
            filters=filt)[:] = colnames

def read_sparse_hdf(handle, group):
    """
    sparse CSC (compressed sparse colums) matrices via hdf5
    """
    shape = group.get_node(group,'v_datadims')
    return csc_matrix(
        (
            group.get_node(group,'v_data'),
            (
                group.get_node(group,'v_indices'),
                group.get_node(group,'v_indptr')
            )
        ),
        shape=shape)
    # if colnames:
    #     handle.create_carray(group,'v_col_names',
    #         atom=tables.StringAtom(
    #             max([len(n) for n in colnames])
    #         ), shape=(len(colnames),),
    #         title="col names",
    #         filters=filt)[:] = colnames

serialization.R

library(rhdf5)

load.sparse.hdf = function (filename, path) {
  idx = as.vector(h5read(filename, paste(path, "v_indices", sep="/")))
  idxptr = as.vector(h5read(filename, paste(path, "v_indptr", sep="/")))
  vals = as.vector(h5read(filename, paste(path, "v_data", sep="/")))
  dims = as.vector(h5read(filename, paste(path, "v_datadims", sep="/")))
  col.names = h5read(filename, paste(path, "v_col_names", sep="/"))
  
  data = sparseMatrix(
    i=idx,
    p=idxptr,
    x=vals,
    dims=dims,
    index1=F
  )
  colnames(data)=col.names
  return(data)
}