arrays - Python - Principal component analysis (PCA) error -

i trying make principal component analysis (pca) using python. here code:

import os pil import image import numpy np import glob matplotlib.mlab import pca  #step1: put database images 3d array filenames = glob.glob('c:\\users\\karim\\downloads\\att_faces\\new folder/*.pgm') filenames.sort() img = [image.open(fn).convert('l') fn in filenames] images = np.dstack([np.array(im) im in img])      # step2: create 2d flattened version of 3d input array d1,d2,d3 = images.shape b = np.zeros([d1,d2*d3]) in range(len(images)):   b[i] = images[i].flatten()  #step 3: pca results = pca(b) results.wt 

but getting error runtimeerror: assume data in organized numrows>numcols

i tried replacing b = np.zeros([d1,d2*d3]) b = np.zeros([d2*d3, d1]) got valueerror: not broadcast input array shape (2760) shape (112)

can me?

if change b = np.zeros([d2*d3, d1]) should change loop afterwards otherwise try put d1 dimention array d2*d3 one.

you should rid of second error doing

you can transpose b

# step2: create 2d flattened version of 3d input array d1,d2,d3 = images.shape b = np.empty([d1,d2*d3])  #if know filling whole array it's faster using np.zeros or np.ones i, im in enumerate(images):      b[i,:] = im.flatten()  #step 3: pca results = pca(b.t) 

i've substituted loop think better version: in implementation first find dimension of images, create list of integers loop on , re-access images. enumerate returns iterator couple (index, value). advantages returns elements need, , don't have access images directly in loop.

probably don't need create images, don't know pil, there can't you. in case, can dimensions like

d1,d2,d3 = len(img), img[0].shape 

---

edit

you if want can convert content of files numpy when reading them.

for records, numpy.asarray.