In this assignment, you will apply decision trees to recognizing digits from images of MNIST handwritten digits. The dataset is included within sklearn (and also R if you want to use R). You will load the dataset and evaluate the decision tree classifier based on cross validation. Below, I will describe the steps using sklearn (python). If you want to use R instead of python, try to implement the same as best as possible given the APIs in R.

Step 1: Load the data: Use the following package from sklearn.datasets import load_digits. Each instance is 8x8 image (64 pixels/features) and there are 1700+ instances.

Step 2: Perform 5-fold cross validation. Use from sklearn.tree import DecisionTreeClassifier, from sklearn.model_selection import cross_val_score to perform cross validation. Vary the max_depth parameter in DecisionTreeClassifier from 1 to 10. How does the average cross validation error change for each of these depths? You can draw a table (or plot a graph using import matplotlib.pyplot as plt).

Step 4: Analyze the feature importances (for the best max_depth setting) (you an refer to the video and demo code). Which were the 10 most important pixels for the classifier. Were the important pixels meaningful to you?

Step 5: Suggest some steps you could take that could improve the performance of the classifier.

You can submit your notebook (or other code file) along with a pdf that presents you analysis. If you don't want to submit a pdf, you can place your analysis within the notebook itself (you can write text in the notebook).