Linear Regression analysis

This is a project comparing the popular linear regression methods Ordinary Least Squares, Ridge and Lasso Regression. All the methods are evaluated first on the famous Franke function, and then on real data from https://earthexplorer.usgs.gov/.

Build with

scikit-image 0.13.1
scikit-learn 0.19.2
pytest 3.2.1
python 3.6.2
numpy 1.13.3
matplotlib 2.0.2
tqdm 4.23.4
imageio 2.3.0

Structure of the repo

Documentation

The documentation of class Lin_Reg contained in cls_reg.py was generated using sphinx.
View documentation by typing
open doc/_build/html/index.html
in the terminal

Src

CV.py: Running cross validation and bootstrap on franke function
CV_terrain.py: Running crossvalidation and bootstrap on real data
Bias_variance_ols.py: Examine the bias vairance trade of
ols.py: Generating plot of mse vs complexity for ordinary least squares
lasso.py: Generating heat map of mse as a function of noise and lambda
ridge.py: Generating heat map of mse as a function of noise and lambd
run_real_data.py: Generating plots of mse vs degree and mse vs
misc_programs: Directory contains programs used to generate example plots cls: Package

• split_patches.py:* Used to split terrain data into random patches of a given size
• plotparams.py: Parameters for ploting
• cls_reg.py: Class containing linear regression methods ols, ridge and lasso; resampling k-fold and bootstrap; statistics MSE and R2
• unit_test.py: A few tests to make sure Lin_Reg class is working
• Franke.py: Generating data from franke function. Also plot if explicitly run.

Data

Contains different terrain data. SRTM_data_Norway_1.tif has been used in this project

Plots

Folder of plots generated. Several not used directly in the project.

Running the tests

Run unit_test.py with pytest -v
pytest unitest.py -v

Authors

• Tommy Myrvik - (https://github.com/myrvixen)
• Kristian Tuv - (https://github.com/kristtuv)