mnist

Open University Computer Vision 22928- Ex1 Classical Machine Learning Algorithms with MNIST.

Full report can be found under /report/report.docx

KNN

Calculate the performance of the KNN classifier, for k=1..10.

PCA

• Calculate PCA on MNIST.
• Draw the average digit, and the 6 first principal components.
• Draw a graph of the explained variance, by n most significant components.
• How many components are required to get to 95% variance? 80%?
• Project the digits to dimension 2, and draw the obtained vectors, each digit with its own color.
• Repeat the KNN question, where each digit is represented by its projection to dimension 2, 10, 20.
• For some digit, project it to dimension k, then restore it. Draw the restoration for different k.
• Calculate PCA for every digit separately. Present the 1st 6 principal components of each.
• Calculate the projection of each test set image to each model.
• Restore by each model.
• Calculate the distance from each restoration to the original image.
• Select the model for which the distance is smallest.