python_tricks.txt

import scipy.io
scipy.io.savemat('sa1.mat', mdict={'arr': mat})
scipy.io.savemat('gmms.mat', mdict={'arr': gmms_})
# export to matlab

components = np.exp(np.einsum('ik...,...km->i...', x_minus_mus[:,:,0],  
        np.einsum('ik...,jk...', inv_sigmas, x_minus_mus)))
# the first einsum is way too slow

import code
code.interact(local=locals()) # gives a REPL

@profile # works with kernprof.py -l -v

import multiprocessing
pool = multiprocessing.Pool()
pool.map(your_function, your_list)
# easy parallel map, `your_function` has to be pickle-able, for more info see
# http://docs.python.org/2/library/multiprocessing.html#using-a-pool-of-workers