Radiant Earth Spot the Crop Hackathon by #ZindiWeekendz
Based on time-series of Sentinel-1 satellite images can you classify crop types in South Africa?
Prize
$300
Time
Ended over 1 year ago
Participants
21 active · 83 enrolled
Helping
South Africa
Forecast
Number of IDs in Starter Notebook and Submission File
Data · 3 Jul 2021, 08:55 · edited 24 minutes later · 9

From running the starter notebook, there are 35,295 IDs in the test set. In the Sample Submission, there are 35,390 IDs, which is used to score the leaderboard. Can you please look into this?

Also, the number of submissions per day is still 10. Is this still the allowed limit for a 3-day hackathon?

Discussion 9 answers

I've pasted the missing field_ids below. What is everyone else doing to work around this? Putting them in the df with probs of 0 gets a bad score... But surely we can only guess about the fields below??

1504, 1544, 2613, 2747, 5123, 5813, 5893, 6051, 9384, 11510, 13408, 14534, 14652, 20502, 20791, 23273, 23428, 24440, 24586, 24844, 27550, 28366, 29607, 30101, 31093, 31121, 34576, 38466, 39273, 39903, 42255, 42862, 43400, 44399, 46830, 46831, 48390, 50379, 50561, 53686, 55156, 56280, 56822, 57008, 59078, 60581, 63020, 64002, 64807, 64989, 65038, 65804, 67244, 71551, 72075, 75407, 75709, 76235, 77452, 78113, 79396, 80161, 80726, 82251, 82904, 83109, 84131, 84518, 85427, 86358, 86742, 91681, 92086, 93155, 93725, 94169, 94861, 95252, 96629, 96746, 99478, 100120, 102026, 102162, 103802, 104605, 107647, 108168, 109552, 109825, 111757, 112178, 113509, 121814

Yes, predicting with probs of zero gives a bad score. The organisers need to remove the missing IDs from the scoring.

Pleae could you tell me how you are dealing with the missing IDs?

Presently, nothing. The score on the leaderboard is all zeros.

where did you get the starter notebook?

https://github.com/radiantearth/mlhub-tutorials/tree/main/notebooks/South%20Africa%20Crop%20Types%20Competition

Hi Everyone,

We've investigated this issue and found that these missing fields were less than 1 pixel in size and in the dataset generation process were dropped from the raster files. We've updated the sample_submission.csv and field_info_test.csv files to remove these missing fields. If you re-download the label collections you should now have these updated files.

Hi Everyone,

Thanks @RadiantEarth for the clarification. I think that @Zindi should also update the SampleSubmission of the Hackaton, the competitions and update the leaderboard scores ...