gistfile1.txt

# install the irlb python package, in a shell do
# pip install -e git+https://github.com/bwlewis/irlbpy.git#egg=irlb
# Note that I don't expect the python implementation to be as polished as the R version (even though it is by the same author)
# For now it is easier for me to work with b/c I know Python better than R.

from scipy.io import mminfo,mmread
import numpy as np

X = mmread("matrix.mtx")

%timeit S = irlb.irlb(X, 10)
# 1 loop, best of 3: 27.9 s per loop

M = X.tocsr()

%timeit S = irlb.irlb(M, 10)
# 1 loop, best of 3: 8.13 s per loop

## Some info:

# sparsity
In [12]: 1.0 * X.nnz / (X.shape[0] * X.shape[1])
Out[12]: 0.023389614034593154

mu = X.mean(axis=0)

# In [14]: mu.min()
# Out[14]: 0.014661860834504188

# In [15]: mu.max()
# Out[15]: 0.47895412059380754

v = np.random.rand(X.shape[1])

# In [17]: %timeit X.dot(v)
# 10 loops, best of 3: 59.8 ms per loop

# In [19]: %timeit M.dot(v)
# 10 loops, best of 3: 35.6 ms per loop