eikedehling/ml-hackathon

Tutorial, scripts and data for machine learning hackathon

Machine learning hackathon

In this machine learning hackathon we'll explore a dataset about cycle sharing, first part of the day is going to be exploratory analysis, second part of the day we will do modelling and predictions.

In this hackathon we're using jupyter notebooks running in a docker container, so you'll need docker installed on your laptop. The dataset is included in the repository and jupyter environment.

Getting started

1. Clone the ml-hackathon directory to your machine

   git clone [email protected]:EikeDehling/ml-hackathon.git

2. The docker/jupyter environment is fired up using a small shell script:

   $ ./start.sh

   ...

   Copy/paste this URL into your browser when you connect for the first time,
      to login with a token:
         http://localhost:8888/?token=575a53d5c6c8256093550b65c0f24777fe427986143e55d8

3. The script outputs some startup info and finally a link with token, with which you can access the notebook environment. You should open that link (click or copy/paste) in your webbrowser.

4. After opening the notebook environment, you'll see a folder list. Go into the "work" folder and you'll see an example notebook and the dataset. Start by checking out the example notebook, then start trying out your own data analysis ideas. (You can create a duplicate if you want)

Libraries

The main libraries we're using in this hackaton are pandas, seaborn and sklearn. If you want to look at documentation or read up on details, see here:

• Pandas (http://pandas.pydata.org/) which is best described as data-handling library, a python DataFrame
• Seaborn (https://seaborn.pydata.org/) which is a data visualization library
• Scikit-learn (http://scikit-learn.org/) the python machine learning library

Dataset

The dataset in this hackathon is from a cycle sharing project in seattle. It includes information on bicycle stations, trips and weather info. The data is provided as CSV files.

The stations data can be slightly ambiguous, so we'll list the fields and their explanation here:

• station_id: station id
• name: name of station
• lat: station geo latitude
• long: station geo longitude
• install_date: date station was placed in service
• install_dockcount: number of docks on the installation date
• modification_date: most recent date station was modified
• current_dockcount: number of docks on 8/31/2016
• decommission_date: date station was decommissioned

Source: https://www.kaggle.com/pronto/cycle-share-dataset

Inspiration

Some ideas worth exploring:

• What is the most popular bike route?
• How are bike uses or routes affected by user characteristics, station features, and weather?

Feedback

Please let us know how you liked the Hackaton: http://bit.ly/2k4b9TH