For my own study (machine learning algorithms and Python study), I will implement the methods published in PRML in Python.
As a general rule, in the algorithm part, in addition to the Python standard library, ** only Numpy can be used **. We do not use machine learning packages such as scikit-learn or tensorflow. Packages that illustrate results, such as matplotlib, will be used unless they are relevant to the algorithm implementation. I also occasionally use other packages such as scipy when needed (already used for digamma functions, etc.). However, we don't use optimization tools (such as scipy.optimize or tensorflow's automatic differentiation feature) that make implementation significantly easier.
Basically, we will implement one method for each chapter. ~~ If you finish all the steps, you may enter the second lap. ~~ It's for my own study, so I won't explain everything in detail. I will omit the explanation that I decided was not necessary to look back on my own.
Links to articles that have already been implemented, and a list of methods that will be implemented.
| PRML | article |
|---|---|
| Chapter 1 Introduction | Bayesian curve fitting |
| Chapter 2 Probability distribution | Student's t distribution |
| Chapter 3 Linear Regression Model | Evidence approximation |
| Chapter 4 Linear Discriminative Model | Bayesian logistic regression |
| Chapter 5 Neural Network | Error back propagation、Mixeddensitynetwork |
| Chapter 6 Kernel method | Gaussian process regression |
| Chapter 7 Kernel Machine with Sparse Solutions | Related vector regression |
| Chapter 8 Graphical Model | Product sum algorithm |
| Chapter 9 Mixed Model and EM | Maximum likelihood estimation of mixed Gaussian distribution |
| Chapter 10 Approximate Reasoning | Variational mixed Gaussian distribution |
| Chapter 11 Sampling Method | Markov Chain Monte Carlo |
| Chapter 12 Continuous Latent Variables | Bayesian principal component analysis |
| Chapter 13 Series data | Maximum likelihood estimation of hidden Markov model |
| Chapter 14 Joining Models | Conditional mixed model |
Recommended Posts