Tom Effland

This is for future reference, for anyone working off of this model's edge score outputs. There is a semantic naming bug in the dependency decoder that actually switches the `head_arc_representation`...

where do you get candidate entities for a mention from?

7

I'm curious how you compute the candidate entities for some mention. In the paper it says you use a lookup table, but does not say how that lookup table is...

Hi, great work! Would it be possible to host or share the trained model weights from the paper? This would be very helpful for others to build on your work...

something like ``` def get_nan_check_hook(name): def check_grad(grad): if torch.isnan(grad).any(): printf('NaN gradient in {name}') return check_grad ```

Overload the `monitoring` function to accept an int, which sets a monitoring frequency of a function instance that takes the training step int and returns a bool for custom monitoring...

Things that would be good for the walkthrough: * Show the full script as a single code block at the end * Fix typos * Break up longer paragraphs *...

allow a whitelist and blacklist of param names/regexes that should be the monitored set/not monitored set. further allow for a it to be a configuration dictionary: ``` param_name:{ value:True, grad:True,...

Nonprojective dependency entropy

9

Hi Sasha, Wanted to contribute an efficient implementation of entropy for non-projective trees I'm using for a current project. Please let me know of any changes you'd like to see....

Value expectation and 1st order CKY

4

Changes are: * Add 1st order cky implementation with suggested updates * Add value-expectation semiring and test * Make marginal computation work on potentials w/o gradients * Minor clarity/consistency changes...