Practical Machine Learning

We offer a foundation in building intelligent business applications using machine learning, walking you through all the steps to prototyping and production—data cleaning, feature engineering, model building and evaluation, and deployment—and diving into an application for anomaly detection and a personalized recommendation engine. All concepts will be presented with example code in Python.

All students should be come prepared with a laptop, in order to follow along with the presentations and to work exercises on their own. It is best to have your laptop configured with Python and the data downloaded before the workshop starts.

This repo will be updated with example code and exercises in the coming weeks. Please remember to pull again before the conference starts.

Installation

All of the code we present uses Python 2.7. A number of libraries beyond the standard library are used. We recommend using the conda package manager to install the same versions that we are using, in a manner that won't interfere with your system packages. Alternatively, you can create a docker container from the provided Dockerfile that will contain all the necessary Python modules.

Clone the repo

$ git clone https://github.com/thedataincubator/strata-practical-ml.git

Conda Option

You may install either the full Anaconda Package Manager or the smaller Miniconda system. The former will provide you with over 720 packages, ready to use; the latter will makes it easy to download them when needed. Once one of these are installed, you can install the packages we will be using into a separate environment with

$ conda env create -f environment.yml

This will create a new conda environment named strata. It can be activated on Linux and OS X with

$ source activate strata

or on Windows with

> activate strata

Docker Option

If you run into trouble with the above installation, consider using Docker.

First, install Docker on your system. Then run these commands

docker build .

This will build a docker image from Dockerfile and automatically run conda installation in a "virtual" Linux environment. Then you can launch the IPython Notebook Server inside of the container with

docker run -p 8888:8888 -v <absolute local path to this directory>:/mnt/local -t -i <image-name>

The notebooks should be available at http://localhost:8888/. The specified directory will be mounted in the container at /mnt/local, which is being served by the Notebook Server.

A shell inside of the container can be obtained with

docker run -p 8888:8888 -v <absolute local path to this directory>:/mnt/local -t -i <image-name> /bin/bash

Note: this has been tested on OS X and Ubuntu Linux.

Data

All of the material will use real-world data sets. We recommend that you download them to your personal machine before the day of the workshop. Two applications will be presented.

Recommendation Engine

We will be using the MovieLens 10M data set, assembled by the University of Minnesota. The data are available in a single 63 MB zip file, available at http://files.grouplens.org/datasets/movielens/ml-10m.zip. Download the zipfile into the recommendation directory, and unzip as follows:

$ cd recommendation
$ wget http://files.grouplens.org/datasets/movielens/ml-10m.zip
$ unzip ml-10m.zip

Anomaly Detection

We will be using data from the New York CitiBike program. This is available in a number of zip files at https://s3.amazonaws.com/tripdata/index.html. They can be easily downloaded with the provided script:

$ ./download.sh

If you have installed the docker container, the script can be run like so:

docker run -v <absolute local path to this directory>:/mnt/local <image-name> /bin/sh -c 'cd /mnt/local && ./download.sh'

Discussion

You can join our gitter room to discuss the projects with your fellow students.