Re-training vs training online

I finally was able to get a basic machine learning model trained on the tournament data (using TensorFlow). I would like to know if I should train the model again when the next data set is released or if I should use the already trained model and use “online” training to feed it more data.

Should my model be discrete or continuous? Has anyone found better/worse performance from week to week from completely re-training?

The training and validation data doesn’t change week-to-week. (Once in a while new validation data is added – you’ll see an announcement in rocketchat when that happens.) So there is no reason to re-train all the time. The only new data each week is the “eraX,live” data which we are actually scored on for the round.