Submissions are scored on the mean of the 50th and 95th percentile distance errors. For every `phone` and at every `millisSinceGpsEpoch`, the horizontal distance (in meters) is computed between the predicted lat/lng and the ground truth lat/lng. These distance errors form a distribution from which the 50th and 95th percentile errors are calculated (i.e. the 95th percentile error is the value, in meters, for which 95% of the distance errors are smaller). The 50th and 95th percentile errors are then averaged for each phone. Lastly, the mean of these averaged values is calculated across all phones in the test set.

## Submission File
For each `phone` and `millisSinceGpsEpoch` in the test set, you must predict the latitude and longitude. You must provide estimations for all timestamps that have at least four valid GNSS signals. The sample submission contains a list of these timestamps. The submission file should contain a header and have the following format:

```
phone,millisSinceGpsEpoch,latDeg,lngDeg
2020-05-15-US-MTV-1_Pixel4,1273608785432,53.599227001298125,-2.4339795741464334
2020-05-15-US-MTV-1_Pixel4,1273608786432,53.599227001298125,-2.4339795741464334
2020-05-15-US-MTV-1_Pixel4,1273608787432,53.599227001298125,-2.4339795741464334    
```
etc.