bow2adjlist.py

#!/usr/bin/env python

"""
  bow2adjlist.py
  
  Convert a sparse matrix (eg term-document matrix) to two dense matrices
  
  Useful for feeding into BOW models
"""

import numpy as np
from scipy.sparse import coo_matrix

def bow2adjlist(X, maxcols=None):
    x = coo_matrix(X)
    _, counts = np.unique(x.row, return_counts=True)
    pos = np.hstack([np.arange(c) for c in counts])
    adjlist = csr_matrix((x.col + 1, (x.row, pos)))
    datlist = csr_matrix((x.data, (x.row, pos)))
    
    if maxcols is not None:
        adjlist, datlist = adjlist[:,:maxcols], datlist[:,:maxcols]
    
    return adjlist, datlist