fonnesbeck/ngcm_sklearn_2017

scikit-learn course for 2017 NGCM Summer Academy

Machine Learning using scikit-learn

NGCM Summer Academy 2017

Instructors

Christopher Fonnesbeck (Vanderbilt University Medical Center)
Skipper Seabold (Civis Analytics)

Outline

Thursday, June 29

09:30 - 10:45 (Chris Fonnesbeck)

Introduction to machine learning with scikit-learn

• 1.a) Introduction
• 1.b) Data preparation and preprocessing

11:00 - 13:15 (Chris Fonnesbeck)

Unsupervised learning

• 1.c) PCA and transformations
• 1.d) Clustering

13:15 - 14:15 Lunch

14:15 - 16:00

Supervised Learning (Skipper Seabold)

• 2.a) Support vector machines
• 2.b) Decision trees and Random Forests

16:15 - 17:30

Model selection (Skipper Seabold)

• 2.c) Model selection and validation

Friday, June 30

09:30 - 10:45

Supervised Learning (Chris Fonnesbeck)

• 3.a) Linear regression
• 3.b) Logistic regression

11:00 - 13:15 (Chris Fonnesbeck)

Ensemble Supervised Learning

• 3.c) Boosting
• 3.d) Gaussian processes

13:15 - 14:15 Lunch

14:15 - 16:00 (Skipper Seabold)

Advanced topics

• 4.a) Pipelining
• 4.b) Feature selection

16:15 - 17:30 (Skipper Seabold)

Advanced topics

• 4.c) Text feature extraction
• 4.d) Big Data Strategies

Prerequisites

This is an intermediate-level computing course, so some previous experience with Python is required. Some undergraduate-level statistics is also recommended.

Software Requirements

Python 3.5 or 3.6. We recommend installing the free Anaconda distribution of Python, available from Continuum Analytics.

The following packages should be installed on your system:

• jupyter
• ipython>=4.0
• numpy>=1.10
• pandas>=0.18
• scipy
• matplotlib
• scikit-learn
• dask

If you have installed Anaconda, most of these may already be available to you.

Getting this repository

git clone https://github.com/fonnesbeck/ngcm_sklearn_2017.git

If you are not familiar with Git and GitHub, you can simply download the zip file of the repository at the top of the main repository page.

Then, move to the directory created by the clone/zip file:

cd ngcm_sklearn_2017

and install everything using conda:

conda config --add channels conda-forge
conda env create -f environment.yml

This will create an environment called sklearn that includes the packages required for the course.
​
If you are not using the Anaconda Python distribution, you will need to manually install the packages listed in environment.yml using pip.

Which you probably don't want to do.

So install Anaconda.

To use the environment, you may type:

source activate sklearn

ngcm_sklearn_2017's Issues

Finalize examples and hands-on exercises

We need to identify datasets and examples for use in the class. I've added ones that I've used in my graduate class, but they may not be optimal.

A good recurring case study or two might be a good idea.

Add attribution and license

I've taken quite a few examples from Andy and the sklearn user guide. Should note this somewhere and include appropriate licenses.

Add additional course content

Here is an outline with a todo list for various sections. Please edit as needed and update status. They can be checked off if they are mostly there, as I'm sure we will tinker until the course starts.

Thursday morning

• Introduction to sklearn API
• Data preprocessing
• Dimensionality reduction
• Clustering

Thursday afternoon

• Support vector machines
• Decision trees
• Random forests

Friday morning

• Linear regression
• Logistic regression
• Model selection
• Non-linear regression
• Boosting
• Regression trees

Friday afternoon

• Pipelining
• Feature selection
• Text feature extraction
• Parallelism with Dask
• Capstone project