Description
Question 1: Principle Component Analysis for Data Compression (30 points)
In our lecture, we have discussed about Principle Component Analysis (PCA), which can be used in
multiple different applications, like data projection for dimensionality reduction.
Here we will use PCA for
image compression and reconstruction. Below is a picture for the Natural Park of Montseny. Can you use
PCA to extract the components of the image and reconstruct the image with the first 100 components, 200,
300 components or even more? And also make a plot for accumulative variance (y-axis) with the number
of principle components (x-axis). (You need to implement by yourself, not calling a function.)
Figure 2: Figure for q1.
Question 2: Naive Bayes (40 points)
During our lecture, we have talked about Naive Bayes. But we didn’t got a change to go deep how to
implement Naive Bayes for a real world application. First I encourage you to read through the additional
learning material with different probability models. Then try to solve the following problem:
People’s name can be connected to which country he/she comes from. Here we have 4000 (fake) names:
Japanese, American, Arabic, and Greek. Implement a NB classifier that can make a prediction given a new
name.
Hint:
• First read through the additional material and think about which probabilistic model you would like
to choose.
• Merge all the names and split them into training (70%) and testing (30%) with shuffle = True.
• You can use CountVectorizer from sklearn.feature extraction.text to vectorize your input names as a
preprocessing step.
• With vectorized representation of your input, then you can implement the algorithm.
Then report your testing accuracy with your algorithm.
