A starter notebook will be provided to help you make your first submission.

The data for this competition takes the form of GPS locations recorded during runs or walks. Each activity is logged by two devices: a smartphone (low accuracy and occasional signal loss) and a ‘benchmark’ device with a much more accurate GPS. The goal is to be able to correct the locations reported by the low-accuracy source device when signal is lost.

To evaluate performance on this task, we’ve split the tracks into ‘sequences’. For each sequence, you’ll be given the GPS recordings from the source device from approximately 120 seconds of activity. Using the source data you must predict the benchmark points for the same 120 seconds.

We have prepared some training sequences in the same format as the test sequences, with the labels available in Train.csv.

We have also shared the raw data used to generate these training sequences. These tracks consist of longer recordings of entire activities by the source and benchmark devices, which can be used to generate additional training sequences. The starter notebook shows examples of how to slice these tracks into sequences similar to those used in the test set.

A sequence is encoded as a CSV file with the following columns:

• Time - Time in seconds from the start of the sequence.
• Latitude and Longitude - the location at that time according to the device. Note that these locations have been shifted to center roughly on (0, 0) and the entire track may have been rotated or otherwise transformed as well.
• Movement_Type - Different kinds of activity are represented as different movement types
• Accuracy - The GPS accuracy according to the device

Files available for download:

• Train.csv - Labels for the training sequences
• Train_sequences.zip - the actual sequences whose labels are stored in Train.csv. The name indicates the track from which the sequence was derived
• Train_tracks.zip - The full recordings from the source and benchmark devices which can be used to generate additional training data.
• Test_sequences.zip - Contains 445 test sequences, each with a unique ID. You must submit predictions based on each of these sequences following the sample submission format
• SampleSubmission.csv - shows the submission format for this competition.
• Starter_Notebook.ipynb - Shows the data format, contains some utilities for plotting sequences, reading in the tracks and making a submission.

Check the starter notebook for additional explanation, and feel free to ask for clarification on the discussions page.