Data Science

Some general machine learning exercises one can run to get used to python's scientific computing libraries.

Getting the most from this project

First off, read this README!

In any given script, there are a lot of moving components flying around that are important to understanding the modeling and evaluation process. If you're new to modeling, here are some basic questions you could ask yourself:

• What is classification? What is regression?
• What is the UCI machine learning repository? Why is it used as a benchmark in many papers that introduce machine learning models?
• What is the difference between python pandas, numpy, and scikit-learn?
• How are the models in each of the scripts different from each other?
• What is k-fold cross validation?
• The iris dataset uses accuracy_score to evaluate its models for classification. What is the difference between evaluating a model using AUC and an accuracy score? Why can't we use AUC to evaluate the models trained in the iris dataset?

Missing bits and pieces

You could help contribute to this project! See below for contribution rules. Some popular tasks that are currently missing are:

• Multitask Classification (The iris dataset is the only multi-class problem)
• Clustering (It's hard to get ground truth data sets)
• Text mining (Again, hard to obtain ground truth)
• Computer Vision (Not my area of specialty)

This repository primarily fetches from the UCI machine learning repository. Please cd data to see the fetching scripts, and the URLs for each address.

Installing the required packages

I put everything into a requirements file. Just do

pip install -r requirements.txt

and let the magic of pip take over. If you need to install these packages on a machine you don't have sudo access to, throw in the --user argument.

Contributing to this project

I'd love it if people contributed to this repository. But here are some ground rules, just to make sure the coding style is consistent.

1. Use 4 spaced indents, no tabs.
2. Try to keep lines within 80 characters, even in the README.md. (PEP 8 says 79, I'm a bit more flexible.) If you use vim 7.4, I'd recommend using the set colorcolumn=80 setting.
3. Do not commit data sets. Create simple fetching scripts like I do.
4. Avoid small changes to commits, if possible. (Yes, I'm guilty of doing this too.) Perform a git rebase whenever you need to squash commits together.