Skip to content


Switch branches/tags

Latest commit


Git stats


Failed to load latest commit information.
Latest commit message
Commit time

Bee-o-diversity challenge RAMP

Authors: Mehdi Cherti & Balazs Kegl

Build Status

Getting Started

An Ubuntu 16.04 AMI image named pollenating_insects_users_3 has been made available at the Oregon site of AWS. We used it with g3.4xlarge instances, but it may work with other GPU insances as well.

After launching the instance and logging in, simply run

cd pollenating_insects_3


ramp_test_submission --submission <submission>

A dedicated notebook is avaiable to get you started on the problem.

The offical competition rules are also in this notebook. You accept these rules automatically when you make a submission at the RAMP site.

Ramp overview

The library ramp-workflow contains tools to define data challenges and a script to test submissions. As a participant all you only need to know is that the RAMP workflow loads and test the files in submissions/<submission>/.

For this competition you need to submit two files:

  1. containing your model. It should contain a class implementing fit and predict_proba methods.
  2. It should contain a function named transform and optionally transform_test.

Go to ramp-workflow for more help on the RAMP ecosystem.

Making a submission

To make a submission you first need to sign up to the RAMP site, then sign up to the challenge event. Both sign-ups need approval, so be patient. Once you are approved, you can submit and in your sandbox.

Before making a submission, please check that your code will properly run on the backend by running:



ramp_test_submission --submission <submission>

Experimeting on your own setup

You can also run experiments on your own setups. To do so please do as following.

Bear in mind that your submission we be run on our backend and using non supported libraries will make submission fail.

$ git clone

Downloading data

Download the data (~17GB) by running


the first time. It will create data/imgs and download the images there using the names <id>, where <id>s are coming from data/train.csv and data/test.csv. If images are properly unzipped in data/imgs, you can delete the zip file data/ to save space.

Installing dependencies

The installation script used to make the AMI is also available. Depending on your current installation, you may not need to execute all of this, but it shows the versions of the various libraries against which we tested the starting kit.

Keras channel

You should set image_data_format to channels_last in ~/.keras/keras.json.


Pollenating insects (third) RAMP starting kit






No releases published


No packages published