efficiently Iterating with numpy

Question

I made code that calculates the average value for each element in period p for array y

import numpy as np
p=4

y =np.asarray([146, 96, 59, 133, 192, 127, 79, 186, 272, 155, 98, 219])

c=len(y)/p
print c

a=[]
for i in range(1,c+1):
    s=y[p*(i-1):p*i]/np.mean(y[p*(i-1):p*i])
    a = np.append(a, s)
print a

b=[]
for i in range(c+1):
    s = np.mean(a[i::p])
    b = np.append(b, s)

print b

Is there a more efficient way to do this instead of using append and the for loops? I do not need both arrays just b

To get rid of the append, allocate two arrays of ints using np.zeros(shape), then assign to the relevant index and you should see a performance boost. Also, what's the end goal of this script? That might help us give you a hand... — Madison May
– Madison May, Commented Jan 1, 2014 at 19:21
Probably, second range should be also in range(p), as you iterate modulo p: a[i::p]. I updated answer, last result equals your expected b, but is one item longer, since c=3 but p=4. — alko
– alko, Commented Jan 1, 2014 at 19:34
@MadisonMay adding Holt-Winters exponential smoothing, this is used for calculating the seasons — user3084006
– user3084006, Commented Jan 1, 2014 at 19:37
@alko That is what I am working on. But for Statsmodel. I got all the double and single down with all the multiplicative models and dampening variations too. I will submit it after I finish the triple and the prediction bands. — user3084006
– user3084006, Commented Jan 1, 2014 at 20:06

alko · Accepted Answer · 2014-01-01 19:57:12Z

2

In fact, you do a 2d operation. Your first array, a, can be evalueated if a second demension added (for example with reshape):

z = y.reshape(-1, p)
w = z/z.mean(axis=1).reshape(-1,1)
print w.flatten()
# [ 1.34562212  0.88479263  0.5437788   1.22580645  1.31506849  0.86986301
#   0.54109589  1.2739726   1.46236559  0.83333333  0.52688172  1.17741935]

Your second, b, is a mean of previous result:

print w.mean(axis=0)
# [ 1.37435207,  0.86266299,  0.53725214,  1.2257328 ]

Update

As you mention exponential smoothing in comments, you might be interested in pandas or statsmodels packages to deal with timeseries. See for example pandas docs with some useful computational tools and this ER in issue tracker for some useful links about exponential smoothing implementation.

edited Jan 1, 2014 at 19:57

answered Jan 1, 2014 at 19:26

alko

48.7k12 gold badges99 silver badges105 bronze badges

Sign up to request clarification or add additional context in comments.

2 Comments

user3084006 Over a year ago

I am basing the calculations method from itl.nist.gov/div898/handbook/pmc/section4/pmc435.htm and using the data from adorio-research.org/wordpress/?p=1230 But it seems like I am getting the wrong numbers for S (first 4 numbers for the seasonal index) or the calculation method at adorio-research might be wrong

user3084006 Over a year ago

if y has an remainder can is the best way to coerce it to work in the reshape function to do an if statement then w = z[:len(y) % p*-1].reshape(-1, p)

Collectives™ on Stack Overflow

efficiently Iterating with numpy

1 Answer 1

2 Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

2 Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related