Code for the Kaggle Connectomics challenge
Switch branches/tags
Nothing to show
Clone or download
Pull request Compare This branch is 11 commits behind asutera:master.
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.

Kaggle Connectomics

Team AAAGV (#1)



The following programs and packages were used for the contest:

- Python 2.7
- NumPy >= 1.6.2
- SciPy >= 0.11
- scikit-learn >= 0.14

with appropriate blas and lapack binding such as MKL, accelerate or ATLAS. In order to test the code, we recommend you to use the Anaconda python distribution (

Code ran on MacOsx 10.9.2 and Linux (version 2.6.18-194.26.1.el5 ( (gcc version 4.1.2 20080704 (Red Hat 4.1.2-48)) 1 SMP Tue Nov 9 12:46:16 EST 2010).

How to train your model

No model is learnt to produce the connectivity score matrix.

How to make predictions on a new test set

In order to reproduce the result, you can launch the file. The usage is the following:


optional arguments:
  -h, --help            show this help message and exit
                        Path to the fluorescence file
  -p POSITION, --position POSITION
                        Path to the network position file
  -o OUTPUT, --output OUTPUT
                        Path of the output/prediction file
  -n NETWORK, --network NETWORK
                        Network name

For example on the "test" dataset, you would use the following command:

python -f fluorescence_test.txt -p networkPositions_test.txt -o score_test.csv -n test

To run the script, you will need a machine with at least 8GB RAM, a fast processor (> 2.5 GHz), 4 cores and sufficient disk space. On our last test, it took around 10 hours (with 7 process) on normal-1 and +-2 hours on small-1.

The performance obtained on normal-1 and small-1 are the following:

On normal-1: 0.94356018640593564
On small-1: 0.71027913026472989

Note that all parameters have been optimized for a big dataset, i.e. 1000 neurons, and it explains the poor result on small-1.