Pulling pre-generated ECGs for training
Sam J
Are we allowed to pull pre-generated training data from a repository during the official phase (a bit like we are allowed to do so for pre-trained models)? Can we include it with the submission repository?
PhysioNet Challenge
Dear Sam,
We encourage you to generate training data during the training step using the provided ECG image generation code or other image generation code so that your entry reflects your method as closely as possible. However, you can also augment the training data by including it in your submission or by downloading or pulling it with a command in your Dockerfile, with a few caveats:
1. Please be careful not to include sensitive, protected, or identifiable health information, such as unredacted images of paper ECGs.
2. Please document any pre-generation steps, much like you would with a pre-trained model for transfer learning, for the reproducibility and reusability of your work.
3. Your training and inference code will not have network access, so it will not be possible to retrieve data once we begin to train and evaluate your code. If you cannot include files in your repository for some reason, e.g., large file sizes, then you can retrieve them with a command in your Dockerfile.
4. We do not have an explicit limit for the amount of data, but there are practical limits, such as the size of the repository or disk or the time allotted to building a Docker image. If your code fails due to one of these limits, then please write back so that we can discuss.
Best,
Matt
(On behalf of the Challenge team.)
Please post questions and comments in the forum. However, if your question reveals information about your entry, then please email info at physionetchallenge.org. We may post parts of our reply publicly if we feel that all Challengers should benefit from it. We will not answer emails about the Challenge to any other address. This email is maintained by a group. Please do not email us individually.