This is a golang version of the Machine learning course from caltech: Learning from data.
- week 1:
- PLA (Perceptron learning Algorithm)
- week 2:
- Hoeffding Inequality
- Linear Regression
- Nonlinear Transformation
- week 3:
- Generalization Error
- week 4:
- VC bound
- Bias and Variance
- week 5:
- Linear Regression Error
- Gradient Descent
- Logistic Regression
- week 6:
- Overfitting and Regularization With Weight Decay
- Neural Networks
- week 7:
- Validation
- Estimators
- Cross Validation
- PLA vs. SVM
- week 8:
- Support Vector Machines With Soft Margins
- Polynomial Kernels
- Cross Validation
- RBF Kernel
##Build:
There is a specific directory week<x> for the homework of each week (1 to 8). To build it run the following command where x = 1
go get ./week1
##Run: Similarly you can run the work of a specific week as follows:
week1
##Test: Tests will be slow as they are running the homeworks whom typically have to run multiple "runs" (1000 or more runs) and compute an average.
go test ./week1
##Todo:
- refactor
- concurrent runs.
- command line animations. Pretty command line / console output on Unix in Python and Go Lang
- refactor PLA and other functions into separate packages.
- linear regression should have a Xn array and an Zn collection when a transformation takes place
- add transpose function.
- transformation function should accept array with param x0 = 1 to transform
- better and consistent print statements.
- catch all error and have all functions send errors.
- add tests
##Current tree:
$ tree
.
├── LICENSE
├── README.md
├── biasAndVariance
│ └── biasAndVariance.go
├── data
│ ├── in.dta
│ └── out.dta
├── generalizationError
│ └── generalizationerror.go
├── gradientDescent
│ └── gradientDescent.go
├── hoeffding
│ └── hoeffding.go
├── linear
│ └── linear.go
├── linreg
│ ├── linreg.go
│ └── matrix.go
├── logreg
│ └── logreg.go
├── measure
│ └── measure.go
├── pla
│ └── pla.go
├── week1
│ ├── week1.go
│ ├── week1_test.go
├── week2
│ └── week2.go
├── week3
│ └── week3.go
├── week4
│ └── week4.go
├── week5
│ └── week5.go
├── week6
│ └── week6.go
└── week7
└── week7.go
##Thoughts:
It might be better to divide the packages based on models and methods. Here is how the topics are presented in the learning from data web page: topics
###models:
- linear classification: PLA
- linear regression
- logistic regression
- non linear transformation
- neural networks
- support vector machines
- nearest neighbors
###methods:
- regularization
- validation