For this project, I to applied nonlinear logistic regression to train a classifier of Street View House Numbers(SVHN data set). I based my project off of the NeuralNetworkClassifier from a previous Machine Learning homework assignment, with which I classified the a data set of Handwritten Digits(MNIST data set).
The SVHN data is very similar to the MNIST data set, but the SVHN data comes from a significantly harder, unsolved, real world problem (recognizing digits and numbers in natural scene images). SVHN is obtained from house numbers in Google Street View images
The goal of my project was to analyze how the classification of the MNIST data compares to the SVHN data. I was able to conclude that the two data sets had very different problems with classifications. The digits that the MNIST classifications did poorly on was due to poor penmanship. While the network training the SVHN data performed poorly on different numbers due to various obstructions in the images. The MNIST data had a significantly higher percent correctly classified samples compared to the SVHN data.