Alleen nullen, enen en tweeën (3)

Het is nu ook gelukt om alles goed (redelijk :-) weer te geven. Bijvoorbeeld:

Prediction for number with index: ( 0 ) =  [2] Known value =  2 Prediction for number with index: ( 1 ) =  [3] Known value =  3 Prediction for number with index: ( 2 ) =  [2] Known value =  2 Prediction for number with index: ( 3 ) =  [2] Known value =  2 Prediction for number with index: ( 4 ) =  [0] Known value =  0 Prediction for number with index: ( 5 ) =  [2] Known value =  2 Prediction for number with index: ( 6 ) =  [2] Known value =  2 Prediction for number with index: ( 7 ) =  [2] Known value =  2 Prediction for number with index: ( 8 ) =  [1] Known value =  1 Prediction for number with index: ( 9 ) =  [1] Known value =  1

Het leeralgoritme blijkt dus inderdaad goed te werken. Het geeft de juiste som van de afzonderlijke cijfers weer.
Het is nu natuurlijk ook eenvoudig om gewoon het getal weer te geven:

Prediction for number with index: ( 0 ) =  [20] Known value =  20

Prediction for number with index: ( 1 ) =  [12] Known value =  12

Prediction for number with index: ( 2 ) =  [2] Known value =  2

Prediction for number with index: ( 3 ) =  [11] Known value =  11

Prediction for number with index: ( 4 ) =  [0] Known value =  0

Prediction for number with index: ( 5 ) =  [2] Known value =  2

Prediction for number with index: ( 6 ) =  [11] Known value =  11

Prediction for number with index: ( 7 ) =  [2] Known value =  2

Prediction for number with index: ( 8 ) =  [1] Known value =  1

Prediction for number with index: ( 9 ) =  [10] Known value =  10

Wow! wat knap is ie toch :-)

Gebruikte code:

import matplotlib.pyplot as plt

from mpl_toolkits.axes_grid1 import ImageGrid

import numpy as np

import time

from sklearn import svm

from load import mnist

trX, teX, trY, teY = mnist(onehot=True)

t0 = time.time()

ga = 0.028          # Hiermee (gamma = 0.028) wordt bij 10.000 oefeningen 97.07% betrouwbaarheid gehaald

size = 1000

clf = svm.SVC(gamma=ga, C=100)

# Eerst nullen, eenen en tweeen voor de oefendataset selecteren.

Y = np.argmax(trY,axis=1)

index=[]

elem = Y.size

for x in range(0,elem-1):

    if Y[x] in range(0,3):

        index += [x]

#print Y[x]

XX = np.take(trX, index, axis=0)

YY = np.take(Y,index)

# Nu nullen, eenen en tweeen voor de testdataset selecteren.

Y = np.argmax(teY,axis=1)

index=[]

elem, val = XX.shape

for x in range(0,size):

    if Y[x] in range(0,3):

        index += [x]

XXt = np.take(teX, index, axis=0)

YYt = np.take(Y,index)

#Nu samenstellen

XXX = np.concatenate((XX,np.flipud(XX)), axis = 1)

YYY = YY+np.flipud(YY)

XXXt = np.concatenate((XXt,np.flipud(XXt)), axis = 1)

YYYt = YYt+np.flipud(YYt)

clf.fit(XXX[0:size],YYY[0:size])

#clf.fit(XXX,YYY)     # Totale set 'leren'.

print ""

print "reliability: ", np.mean(YYYt == clf.predict(XXXt)), time.time() - t0, "seconds, gamma: ", ga, " size :", size

print ""

t = 0

i = 0

fig = plt.figure(1,(5.,5.))

grid = ImageGrid(fig,111,

                 nrows_ncols = (10,2),

                 axes_pad=0.1,

                 )

while (t < 19 and i < 100):

    if (YYYt[i] != clf.predict(XXXt[i])):

        print "Prediction for number with index: (",i, ") = ", clf.predict(XXXt[i]), "Known value = ",YYYt[i]

        im1 = XXXt[i,0:784]

        im2 = XXXt[i,784:2*784]

        im1.shape=28,28

        im2.shape=28,28

        grid[t].imshow(im1,cmap=plt.cm.gray_r)

        grid[t+1].imshow(im2,cmap=plt.cm.gray_r)

        #plt.imshow(im1, cmap=plt.cm.gray_r, interpolation="nearest")

        t += 2

    i += 1

plt.show()