Fast Iteration of numpy arrays

Question

I'm new to python and I'm trying to do some basic signal-processing stuff and I'm having a serious performance problem. Is there a python trick for doing this in a vectorized manner? Basically I'm trying to implement a 1st order filter, but where the filter characteristics may change from one sample to the next. If it were just one filter I would use numpy.signal.lfilter(), but it's a bit trickier. Here's the snippet of code that goes very slowly:

#filter state
state = 0

#perform filtering
for sample in amplitude:
    if( sample == 1.0 ): #attack filter
        sample = (1.0 - att_coeff) * sample + att_coeff * state
    else: #release filter
        sample = (1.0 - rel_coeff) * sample + rel_coeff * state

    state = sample

Answer 1

You could consider using one of the Python-to-native-code converters, such as Cython, Numba or Pythran.

For instance, running your original code with timeit gives me:

$ python -m timeit -s 'from co import co; import numpy as np; a = np.random.random(100000)' 'co(a, .5, .7)'
10 loops, best of 3: 120 msec per loop

while annotating it with Pythran, as in:

#pythran export co(float[], float, float)
def co(amplitude, att_coeff, rel_coeff):
    # filter state
    state = 0

    # perform filtering
    for sample in amplitude:
        if sample == 1.0: # attack filter
            state = (1.0 - att_coeff) * sample + att_coeff * state
        else:             # release filter
            state = (1.0 - rel_coeff) * sample + rel_coeff * state
    return state

and compiling it with

$ pythran co.py

gives me:

$ python -m timeit -s 'from co import co; import numpy as np; a = np.random.random(100000)' 'co(a, .5, .7)' 
1000 loops, best of 3: 253 usec per loop

That's roughly a x470 speedup! I expect Numba and Cython to give similar speedups.