What's the best way to incorporate growth stage? I've tried postprocessing with growth stage + predictions, this only gave modest improvements however.
Has anyone tried building separate models per growth stage? if so did this work?
I think 9-ish is feasible, but I haven't quite got there yet. It's also hard to know how true many of those 9-9.5 scores are. It does seem many people have used the leak, or are using it to hide their true score.
I tried including growth stage as a categorical, but only found a very small increase.
Previous I'd been thinking perhaps classifying 0 vs. nonzero extent would be an easier problem within growth stage - the extent distribution is quite different by growth stage.
Hey, two way to use it, as a features or as a label. Given it is available as well in the test set, you should use it as a categorical feature
I used it as a categorical feature but did not improve my score. I'm curious how people are getting 9-ish scores.
I think 9-ish is feasible, but I haven't quite got there yet. It's also hard to know how true many of those 9-9.5 scores are. It does seem many people have used the leak, or are using it to hide their true score.
I tried including growth stage as a categorical, but only found a very small increase.
Previous I'd been thinking perhaps classifying 0 vs. nonzero extent would be an easier problem within growth stage - the extent distribution is quite different by growth stage.
Did you try 0 vs. nonzero extent classification? Did it help?
I tried this previously on the overall dataset. However, I was using the dataset with incorrect extent ( :/ ).
Am planning on retrying this today. Perhaps growth stage could be used here.