Closed irinakhismatullina closed 5 years ago
@irinakhismatullina Did you use the freshly merged neural splitter?
No, is it ready??? I didn't know that
Did anybody check the quality? I would love to play with it, but after that I will probably have to restart all experiments from the zero point to get the best quality with the fresh data...
The quality is the same as in the paper https://arxiv.org/abs/1805.11651
Caution: it requires a GPU to work, otherwise you finish splitting one million by Christmas. So use the ML cluster.
I'm always using the cluster:) So, what's best to do: look at the new splitter right away, or first add everything I've done with the old one?
Add the old stuff first, that's safer.
You can contact @glimow about using the model, that's him who worked on it.
Then this PR can be merged.
Loaded new dataset (1 million identifiers, as before), changes: