Submissions are evaluated by the mean [Intersection over Union](https://en.wikipedia.org/wiki/Jaccard_index) (IoU, aka the Jaccard index) of the ground-truth entry matches and the predicted entry matches. The mean is taken sample-wise, meaning that an IoU score is calculated for each row in the submission file, and the final score is their average.

## Submission File ##

For each place entry `id` in the test set, you should submit a space-delimited list of matching place `id`s. *Places always self-match*, so the list of matches for an `id` should always contain that `id`.

The file should contain a header, be named `submission.csv`, and have the following format:
```
id,matches
E_00001118ad0191,E_00001118ad0191
E_000020eb6fed40,E_000020eb6fed40
E_00002f98667edf,E_00002f98667edf
E_001b6bad66eb98,E_001b6bad66eb98 E_0283d9f61e569d
E_0283d9f61e569d,E_0283d9f61e569d E_001b6bad66eb98
```
You should predict matches for every `id`. For example, if you believe `A` matches `B` and `C`, your submission file should include rows `A,A B C`, but also `B,B A C` and `C,C A B`.