Regularization methods for machine learning

These contents were taugh in summer school RegML 2016 by Lorenzo Rosasco and this GUI in python was submitted as part of final exam.

Total:

All the coded and tested functions are in RegML.py and GUIs code structure is in RegML_GUIv2.1.py

Github Page

PyPi -project

Installation

pip install regml

Opening GUI:

import regml
regml.GUI()

Regularization Methods

• Regularized Least Squares -RLS Referance
• Nu-Method Referance
• Iterative Landweber Method Referance
• Singular Value Decomposition Reference
• Trunctated SVD Referance 1 Referance 2
• Spectral cut-off

Kernal Learning

(Linear, Polynomial, Gaussian)

• Linear: $K(X,Y) = X’Y$
  • 
• Polynomial: $K(X,Y) = (X’Y +1)^p$
  • 

• Gaussian (RBF): $K(X,Y) = exp(-

  X-Y

  ^2/2\sigma^2)$

  • 

K-Fold Cross Validation

GUI

Regularization for Machine Learning

---

Files

1. RegML.py
2. RegML_GUIv2.1.py
3. Getting_Started_Demo.ipynb

Requirments

Following libraries are required to use all the functions in RegML library

1. Python(=2.7)
2. Numpy(>=1.10.4) Numpy
3. Matplotlib(>=0.98) Matplotlib
4. Scipy(>=0.12) Optional -(If you need to import .mat data files) Scipy

Tested with following version

GUI is tested on followwing version of libraries

• Python 2.7 / 3
• Numpy 1.10.4
• Matplotlib 1.15.1
• Scipy 0.17.0

Getting starting with GUI

Windows————————

After lauching python, go to directory containing RegML.py and RegML_GUIv2.1.py files and run following command on python shell

>> run RegML_GUIv2.1.py

If you are using Spyder or ipython qt, browes to directory, open RegML_GUIv2.1.py file and run it

Ubuntu/Linux——————-

Open terminal, cd to directory contaning all the files and execute following command

$ python RegML_GUIv2.1.py

if you have both python 2 and python 3

$ python2 RegML_GUIv2.1.py

If you are using Spyder or ipython qt, browes to directory, open RegML_GUIv2.1.py file and run it

Getting Started with DEMO

Getting_Started_Demo is a IPython -Notebook, which can be open in Ipython-Notebook or Jupyter

Notebook

RegML Library

---

Cite As

@software{nikesh_bajaj_2019_2646550,
  author       = {Nikesh Bajaj},
  title        = ,
  month        = apr,
  year         = 2019,
  publisher    = {Zenodo},
  version      = {0.0.2},
  doi          = {10.5281/zenodo.2646550},
  url          = {https://doi.org/10.5281/zenodo.2646550}
}

Nikesh Bajaj

n.bajaj@qmul.ac.uk

nikesh.bajaj@elios.unige.it

http://nikeshbajaj.in