Digit Prediction using Multinomial Naive Bayes in Python

The full code can be found here. github

We import the data:

import numpy as np 
import pandas as pd 
import seaborn as sns
import matplotlib.pyplot as plt
%matplotlib inline

train = pd.read_csv(‘train.csv’)
test = pd.read_csv(‘test.csv’ )

Goal is to Classify handwritten digits using the famous MNIST data. The data is in 28X28 pixel data(height and width) and each data point is weighted in integer between 0 and 255. We will convert these data to single array of 784 pixels or data. Visually, if we omit the “pixel” prefix, the pixels make up the image like this:

000 001 002 003 ... 026 027
028 029 030 031 ... 054 055
056 057 058 059 ... 082 083
 |   |   |   |  ...  |   |
728 729 730 731 ... 754 755
756 757 758 759 ... 782 783

We can use the following code to print some sample data.

%matplotlib inline
import matplotlib.pyplot as plt

features_train[0].shape
B = np.reshape(features_train[0], (28, 28))

plt.imshow(B,cmap=plt.cm.gray_r,interpolation=’nearest’)

We extract the features and label of the data.

# Select the first 2 columns
features_train = train.iloc[0:2000,1:785].as_matrix(columns=None)
labels_train = train.iloc[0:2000,0:1].as_matrix(columns=None)

features_test = train.iloc[2001:3000,1:785].as_matrix(columns=None)
labels_test = train.iloc[2001:3000,0:1].as_matrix(columns=None)

We then apply the multi-nomial Naive Bayes.

from sklearn.naive_bayes import MultinomialNB
clf = MultinomialNB()
clf.fit(features_train, labels_train.ravel())

Finally, we apply the prediction to test data.

y_pred = clf.predict(features_test)
y_true = labels_test.ravel()

We evaluate the accuracy of our model.

from sklearn.metrics import accuracy_score
accuracy_score(y_true, y_pred)

which come to be around 0.83283283283283283. Not bad, but lots of opportunity for tuning.