Datasets: Training, Validation and Testing

[bkowshik]

1. Training

• Labelled changesets from osmcha between January, 2017 to April, 2017
• Model will initially be trained on 20% of this dataset called the sample
• Before publishing, model will be trained on 100% of this dataset

2. Validation

• To estimate how well your model has been trained
• Using labelled changesets from osmcha from May, 2017

3. Testing

• All changesets from OpenStreetMap on 1st May, 2017

[bkowshik]

We have the 3 datasets now on s3:

• Training attributes
• Validation attributes
• Testing attributes