hbcbh1999 / cs229-ps

Stanford Machine Learning course CS229(2017) Problem set answers

Problem set solutions to Standford CS229(2017) Machine Learning Open Course by Andrew Ng
The course page: http://cs229.stanford.edu/
The problem set is from: http://cs229.stanford.edu/ps/
This is for personal interest, you are welcom to use the material but keep in mind this is not guaranteed to be the right answer.
It is also really appreciated if you could point out anything wrong or provide any suggestions on improvement of my answer.

Content:
PS1:
1.1 Logistic regression
1.2 Poisson regression and the exponential family
1.3 Gaussian discriminant analysis
1.4 Linear invariance of optimization algorithm (Newton's method)
1.5 Regression for denoising quasor spectra (a step forward of LWS)

PS2:
2.1 Logistic Regression: Training stability
2.2 Model Calibration
2.3 Bayesian Logistic Regression and weight decay
2.4 Constructing kernels (Support Vector Machine, Mercer's theorem)
2.5 Kernelizing the Perceptron (Learning Theory, Kernels)
2.6 Spam classi cation (SVM)

PS3:
3.1 A simple neural network
3.2 EM for MAP estimation
3.3 EM application
3.4 KL(Kullback-Leibler) divergence and Maximum Likelihood
3.5 K-means for compression

PS4:
4.1 Neural networks: MNIST image classification
4.2 EM convergence
4.3 PCA (Principle Component Analysis)
4.4 ICA (Independent Component Analysis)
4.5 MDP (Markov Decision Processes)
4.6 Reinforcement Learning: the inverted pendulum