I have been adapting a neural network to classify images from PyBrain's tutorial:
http://pybrain.org/docs/tutorial/fnn.html
It is feed in image data in png form, each image is assigned a particular class.
It works well until:
out = fnn.activateOnDataset(griddata)
The message it returns is: AssertionError: (3, 2)
Am pretty sure it is a problem with how I have declared the griddata dataset but I don't know exactly what?
On the tutorial version it runs fine.
My code:
from pybrain.datasets import ClassificationDataSet
from pybrain.utilities import percentError
from pybrain.tools.shortcuts import buildNetwork
from pybrain.supervised.trainers import BackpropTrainer
from pybrain.structure.modules import SoftmaxLayer
from pylab import ion, ioff, figure, draw, contourf, clf, show, hold, plot
from scipy import diag, arange, meshgrid, where
from numpy.random import multivariate_normal
import cv2
from pyroc import *
#Creates cover type array based on color of pixels in roadmap
coverType = [(255,225,104,3), #Road
(254,254,253,0), #Other
(254,254,254,3), #Road
(253,254,253,0),#Other
(253,225,158,0),#Other
] # have other cover type but sample amount included
coverTypes = len(coverType)
print coverTypes #to count
#Creates dataset
alldata = ClassificationDataSet(3,1,nb_classes=10)
"""Classifies Roadmap Sub-Images by type and loads matching Satellite Sub-Image
with classification into dataset."""
for eachFile in glob.glob('Roadmap Sub-Images/*'):
image = Image.open(eachFile)
fileName = eachFile
newFileName = fileName.replace("Roadmap Sub-Images", "Satellite Sub-Images")
colors = image.convert('RGB').getcolors() #Finds all colors in image and their frequency
colors.sort() #Sorts colors in image by their frequency
colorMostFrequent = colors[-1][1] #Finds last element in array, the most frequent color
for eachColor in range(1,151): #151 number of element in CoverType array
if colorMostFrequent[0] == coverType[eachColor][0] and colorMostFrequent[1] == coverType[eachColor][1] and colorMostFrequent[2] == coverType[eachColor][2]:
print newFileName #Check new route
image = cv2.imread(newFileName)
meanImage = cv2.mean(image) #Take average color
meanImageRGB = meanImage[:3] #Converts to RGB scale, excluding "alpha"
print meanImageRGB #Check RGB average colors
alldata.addSample(meanImageRGB,coverType[eachColor][3])
tstdata, trndata = alldata.splitWithProportion( 0.25 )
trndata._convertToOneOfMany( )
tstdata._convertToOneOfMany( )
fnn = buildNetwork( trndata.indim, 5, trndata.outdim, outclass=SoftmaxLayer )
trainer = BackpropTrainer( fnn, dataset=trndata, momentum=0.1, verbose=True, weightdecay=0.01)
ticks = arange(-3.,6.,0.2)
X, Y = meshgrid(ticks, ticks)
#I think every thing is good to here problem with the griddata dataset I think?
# need column vectors in dataset, not arrays
griddata = ClassificationDataSet(2,1, nb_classes=4)
for i in xrange(X.size):
griddata.addSample([X.ravel()[i],Y.ravel()[i]], [0])
griddata._convertToOneOfMany() # this is still needed to make the fnn feel comfy
for i in range(20):
trainer.trainEpochs( 1 )
trnresult = percentError( trainer.testOnClassData(),
trndata['class'] )
tstresult = percentError( trainer.testOnClassData(
dataset=tstdata ), tstdata['class'] )
print "epoch: %4d" % trainer.totalepochs, \
" train error: %5.2f%%" % trnresult, \
" test error: %5.2f%%" % tstresult
out = fnn.activateOnDataset(alldata)
out = out.argmax(axis=1) # the highest output activation gives the class
out = out.reshape(X.shape)
figure(1)
ioff() # interactive graphics off
clf() # clear the plot
hold(True) # overplot on
for c in [0,1,2]:
here, _ = where(tstdata['class']==c)
plot(tstdata['input'][here,0],tstdata['input'][here,1],'o')
if out.max()!=out.min(): # safety check against flat field
contourf(X, Y, out) # plot the contour
ion() # interactive graphics on
draw() # update the plot
ioff()
show()
