Updates for November 8

[bfhealy]

Pull requests

• Add group_ids to photometry json #148
• Add custom keywords to training script #149
• Update W UMa case in config defaults #150
• Save/load trained models consistently #152
• Make CSV optional in get_features #154
• Fix dmdt inference bug #155

Issues

• Compare DNN and XGB results #151

Going forward

Hyperparameter tuning has begun, and the DNN models are showing good performace on most classes. The ids -> features -> inference pipeline is being debugged to prepare it for large-scale inference once training is complete. Before proceeding to the whole sky, we will compare the new DNN results with those from the XGB algorithm. This will require us to get ids and features for sources in the 20-field test group from the initial paper.