Peter Plantinga
Peter Plantinga
> The obvious downgrade is that it makes recipe PRs more annoying. This is only for recipe PRs that add data to the test samples right? And the change to...
The docs folder could be useful. As of right now, Delight doesn't support those features. If you think they're important, create a new issue and we'll see what we can...
Nimrod does indeed look very similar to Delight. I don't see list comprehensions or some other features, but it looks way more developed. Thanks for the heads up!
No fix needed. One can simply wrap the list comprehension in an array() function, e.g. array({x for x in 0 .. 5})
Actually, the syntax of [x for x in 0 .. 5] would be nice to have.
I'll be honest, I didn't think there was much interest in the language so I haven't done much with it lately. Either you can just go ahead and port to...
If you do translate your code to delight, and decide later you'd prefer to use D, no harm done because delight produces d code directly.
As for how to help, honestly porting code to this language would be a huge help! Just to see how it does on real code, and to discover the bugs...
Training loss most often exceeds validation/test loss due to data augmentation and regularization strategies such as dropout which make the task more difficult.
On second read, are you asking why your loss is different from the official recipe?