Dear EHR DREAM Challenge Participants, The slides and video recording of the Webinar #2 are now available: https://www.synapse.org/#!Synapse:syn20815689 ## Q&A **Q: Are the test and Validation data set balanced?** A: No, the test set maintains the original death/live proportion. **Q: In the synthetic data, there were some subjects, which didn?t row in visit_occurrence. How should we handling such subjects?** A: Every patient in new synpuf data (release soon) will have at least one visit record. **Q: Are temperatures included in measurements validation data set?** A: Yes, you can use the concept id 45771331 from the measurement table. You can find this code in the available_concepts.csv file. **Q: are the exercise data (synthetic data) that we use similar to the feature distribution of real patients?** A: Our synthetic data doesn?t represent any feature information of the real UW data. **Q: what is fast lane? that is practice of submission? can't count submission?** A: The fast lane is a way to test your model on synthetic data. You can submit up to 5 successful submissions per day on the fast lane. **Q: Leaderboard scores based on auc ?, how aupr is used?** A: Yes, it?s based on auc and auprc if there is a tie **Q: If synthetic data is a perfectly random distribution, how should we determine if our model is good? For example, couldn't pretraining be meaningless if the feature of a synthetic patient who dies within six months is really different from the feature of a patient who would actually die?** A: The synthetic data is only for testing that the model will run and create a valid prediction file. the idea is that if your model passes the synthetic data, it should run all the way through on the UW data. **Q: Are there any plans to attach a GPU to the UW node such that we can train deep learning models?** A: Not for this first EHR DREAM Challenge. Future EHR Challenges may provide GPUs. **Q: Sorry I didn't hear about ncats. Is UW the main leaderboard dataset and ncats a practice set? Can ncats be submitted five times every day for two rounds?** A: the fast lane can be submitted to 5 times a day which runs on NCATs infrastructure (synthetic data). The main challenge queue you submit to will actually first submit to NCATs infrastructure as well to first validate your model will run through the synthetic data prior to running on the UW dataset. You can submit up to three times to this queue based on whether or not your submission is valid when ran on UW data. **Q: Until when we can submit (date, time) to have a successful attempt to round2?** A: Round 2 will end on December 9 at midnight PST.