Hello all! As we discussed a few weeks ago, the organizers have decided to make the Avocado model and predictions available to all the participants. You can find the trained models themselves (one for each chromosome), as well as the imputed training and validation tracks, in the file baselines/avocado. The simplest use of these predictions would be a boosting approach, when your model attempts to learn the difference between the Avocado prediction and the truth using some extra information, or some form of clever modeling. These models are trained using the training data, and then subsequently used to make predictions for both the training tracks (in the training folder), and the validation tracks (in the validation folder). Note that there is a separate model trained for each chromosome, meaning that the genomic latent factors are not comparable across models. This differs from the training approach we took in our paper, but shouldn't effect the imputations. We will soon be releasing a second set of models that are trained on both the training and the validation tracks. These models should be used ONLY for making predictions on the held out test set at the end of the round, and NOT for making predictions on the validation set, which is part of the training set for these models. Hopefully this will help even the playing field between people who have massive compute at their disposal (and could simply recreate the Avocado model) and those who don't. Let me know if you have any questions!