CODEWITHSUNDEEP
pythonbiopython
sam
sam

Reputation: 19164

a specific method for numpy, scipy resampling

I have a numpy array. I want to resample that numpy array to a specific value say 10. 

 X = [[  6.99749994  17.76250029   5.01699996]
     [ 10.5150001   18.28000021   4.06300002]
     [ 12.47374988  19.4937501    6.93949986]
     [ 15.38050032  21.92675018   6.68924999]
     [ 17.19525003  19.25349998   7.71924984]
     [ 15.75849962  17.17449951   5.07899988]
     [ 16.83874989  19.46924973   2.56125003]
     [ 20.24999952  19.40649986   3.77824998]
     [ 20.32649994  15.83099985   3.59350002]
     [ 19.17724943  15.48849988   0.23099999]
     [ 21.44624996  18.01575041  -0.98599999]
     [ 24.13700008  16.26849985   0.35250001]
     [ 23.45549965  13.07250023  -0.88625002]
     [ 22.66449976  14.09524989  -4.22149998]
     [ 25.57133357  15.34866651  -4.50566673]
     [ 27.54475021  12.71549988  -4.02225   ]
     [ 25.7732501   11.273       -6.82424998]
     [ 26.65899976  13.21299966  -9.15133333]
     [ 27.97424984  12.19199991 -12.32075   ]
     [ 26.63675022   8.88499999 -11.40549994]]

Right now my numpy array in 3x20 dimensions. I want to resample to 3x10. How can I do that using Biopython, numpy or scipy?

this is resampling function which I need to convert to python. pastebin.com/JsGeNyLp where input is numpy array and N is any integer value e.g.10

Upvotes: 0

Views: 747

Answers (1)

Slater Victoroff
Slater Victoroff

Reputation: 21914

While the question is a bit vague, if all you're looking to do is randomly select some number of elements from a numpy array at random, python has a convenient random module that will let you do just that:

import random

resampled_X = random.random_sample(X, 10)

Looking at your linked MATLAB code, it looks like you might be looking for some 1d interpolation in addition to this. Again, hard to say exactly without more details in your question, but scipy in fact has an interp1d function just like MATLAB. You would use it like so:

from scipy.interpolate import interp1d
x = range(30)
y = [i**2 for i in x]
# quadratic can be replaced with linear, cubic, or just a number for polynomial degree
new_function = interp1d(x, y, 'quadratic')
new_function(1.5)
>>> 2.25

After reading through your code I can offer the following translation of it into numpy and scipy code, but it's an extremely unintuitive definition for "resampling"

import numpy as np
from scipy.interpolate import intep1d

def resample(X, N):
    norms = [0] + [np.linalg.norm(X[:,i] - X[:,i-1]) for i in range(1,X.shape[1])]
    cumdel = np.cumsum(norms)/sum(norms)
    solution_space = np.linspace(0,1, N)
    new_function = lambda i: interp1d(cumdel, X[i,:])
    return np.array([new_function(i)(solution_space) for i in range(X.shape[0])]).T

Basically what it looks like this code is doing is this:

  • Constructing a list of the magnitude of changes from one entry in the array to the next. (The norms list)
  • Normalizing that list with a cumsum, so it becomes a list of "percent of the way through all of the changes in this list" (cumdel)
  • Interpolates along each column (of the three in your example) and then sampling from a linspace going from zero to one (solution_space) according to the n added (return line)

Kind of a weird thing to be doing, but that's what it's doing. Let me know if you have any more problems.

Upvotes: 4

Related Questions