Oct. 16, 2019, 7:51 p.m. UTC
Dec. 16, 2019, 2:35 p.m. UTC
There is a limit on the number of submissions you can do in this competition. You can check this limit and when your next submission is allowed on your submission page, (note that if you create multiple users to go around this limit you will be disqualified eventually).
Any valid submission will appear on the leaderboard as long as its score is better than your previous one. At the end of the competition your best ranking submission and the previous one will be re-evaluated and used to assemble the final ranking.
Once you have downloaded
test_data.csv in the data section, you can see that it contains a column named
event_id. Your task is to predict the risk value at the time of closest approach, for each
event_id. Note that, at test time, you do not have access to the data from the CDMs that are within 2 days to the close encounter.
The first column must be called "event_id" and must contain exactly the same ids as the ones in the
test_data.csv (the order does not matter), in order for us to map your predictions to the truth. The number of rows must be equal to the number of events (i.e. one row per event).
The second column must be called "predicted_risk" , which is where you would report the risk predicted by your model, for the corresponding collision event id.
If your submitted file is not compliant to the format described above, an error will be raised and your submission will not be validated nor will appear in the leaderboard. You will receive a message detailing the error, and that should help you fixing it (note that we also provide below an example of a valid sumbission file).
Common situations for which your submission will fail to pass the validation phase and return a score:
id_eventcolumn of your submission does not exactly match those in
id_eventmust be the first column
predicted_riskmust be the second one.