Predicting numbers using sklearn digits dataset - wrong predictions

Question

I want to build a simple digit prediction model.

Therefore I:

1. load in the sklearn dataset
2. Expand the figure sizes from 8*8 to 32*32
3. Teach a SVM using the sklearn digits
4. Predict the new image.

--> The model returns 8 or 1 for most of the test images. Do I have a mistike in the code?

The image is the following:

The code I use is:

def predictimage(file):

import matplotlib.pyplot as plt
from skimage import transform
from PIL import Image
import pandas as pd
import numpy as np
from sklearn import svm
from sklearn import datasets
import PIL.ImageOps


#Load in the query instance


img= Image.open(file)
img=img.convert("L")
img=PIL.ImageOps.invert(img)
img=img.resize((32,32),Image.ANTIALIAS)
imgplot=plt.imshow(img)


query=np.array(img).flatten()
query=(query/16).round()



#Plot query digit
plt.imshow(query.reshape((32,32)))

#Load in the training dataset

digits=datasets.load_digits()
features=digits.data
targets=digits.target





#Expand 8*8 image to a 32*32 image (64 to 1024)
newfeatures=[transform.resize(features[i].reshape(8,8),(32,32))for i in range(len(features))]
newfeatures=np.array(newfeatures).reshape((1797,1024)).round()

#Plot expanded image with 32*32 pixels
for l in range(9):
    ax[1+l].imshow(newfeatures[100+l].reshape((32,32)).round())



#Instantiate, Train and predict    
clf=svm.SVC(gamma=0.001,C=100)
clf.fit(newfeatures,targets)

prediction=clf.predict(query)

plt.show()
return prediction



predictimage(r"C:\...\digit.jpg")

array([8])

Answer 1

You need to indent your code:

from matplotlib import pyplot as plt 
from skimage import transform
from PIL import Image
import pandas as pd
import numpy as np
from sklearn import svm
from sklearn import datasets
import PIL.ImageOps

def predictimage(file):

    #Load in the query instance
    img = Image.open(file)
    img =img.convert("L")
    img =PIL.ImageOps.invert(img)
    img =img.resize((32,32),Image.ANTIALIAS)
    imgplot =plt.imshow(img)


    query=np.array(img).flatten()
    query=(query/16).round()



    #Plot query digit
    plt.imshow(query.reshape((32,32)))

    #Load in the training dataset

    digits=datasets.load_digits()
    features=digits.data
    targets=digits.target





    #Expand 8*8 image to a 32*32 image (64 to 1024)
    newfeatures=[transform.resize(features[i].reshape(8,8),(32,32))for i in range(len(features))]
    newfeatures=np.array(newfeatures).reshape((1797,1024)).round()

    #Plot expanded image with 32*32 pixels
    for l in range(9):
        ax[1+l].imshow(newfeatures[100+l].reshape((32,32)).round())



    #Instantiate, Train and predict    
    clf=svm.SVC(gamma=0.001,C=100)
    clf.fit(newfeatures,targets)

    prediction=clf.predict(query)

    plt.show()
    return prediction