GOTO Vegas

Classify transient sources accurately and efficiently.

Leaderboard

Rank	Time	Branch	Commit	Train Time	Test Time	Transients Found	Transients Missed	False Positives	Score
1	17/07/31 11:43	rol-randomforest	06c55536	23s	2s	335	51	7	0.888
2	17/08/02 11:12	rol-randomforest	343c1a9c	22s	2s	342	0	0	0.885
3	17/08/02 10:55	casey-random-forest	e93e35ee	23s	2s	339	0	0	0.882
4	17/07/31 04:19	baseline	5f91179e	0s	36s	259	127	47	0.7
5	17/07/31 04:10	baseline	165655c4	0s	35s	259	127	47	0.7
6	17/08/02 10:13	baseline	64c14196	0s	0s	207	179	6	0.589
7	17/07/30 07:40	baseline	5f91179e	0s	13s	56	39	56	0.569

Submit a classifier for evaluation

Any member of the GOTO organization on GitHub can submit an entry. First, clone this repository and create a branch with a representative name (e.g., something like <last_name>-<short_description>) and switch to that branch:

git clone git@github.com:goto-obs/goto-vegas.git
cd goto-vegas
git branch casey-random-forest
git checkout casey-random-forest

Now create your classifier by changing the behaviour of the Classifier class in classifier/classifier.py. Specifically, you will want to change the code in the train and classify functions.

Here is the worst kind of classifier, which will never find any transient:

# -*- coding: utf-8 -*-

from __future__ import division, print_function
from .base import BaseClassifier

import numpy as np

class Classifier(BaseClassifier):

    def train(self, predictors, classifications, **kwargs):
        """
        Train the model based on a table of predictors and known classifications
        for objects in a training set.

        :param predictors:
            An :class:`astropy.table.Table` of possible predictors, where the
            number of rows is the number of objects in the training set.

        :param classifications:
            An array of classifications for all objects in the training set.
            This array should have the same length as the number of predictor rows.
        """
        return None

    def classify(self, predictors, **kwargs):
        """
        Classify multiple objects, given the predictors for each object.

        :param predictors:
            A table of predictors (one row per object).

        :returns:
            A single-valued classification for each object.
        """
        return np.zeros(len(predictors))

You can test your classifier locally by running python score.py. If you want to submit your entry to the leaderboard, you will need to commit your changes and push them to GitHub:

git add classifier/classifier.py
git commit -m "Add Random Forest entry"
git push --set-upstream origin casey-random-forest

Your classifier will be run on the test set and scored automatically by Travis CI. Once the classifier has been scored, your entry will (hopefully!) appear on the leaderboard. Otherwise, your classifier might have done a very bad job and not made it in to the top ten. All entries evaluated by Travis CI are available here.

Maintainer

• Andrew R. Casey (Monash)