Slow when alias has large number of candidates

[8key]

Hello,

Thanks for open-sourcing this great proj!

I encounter that Chatette is very slow when alias has large number of candidates even when the training example needed is set to very small.
e.g.,
%intent_name
@[generic-name]

@[generic-name]
~[name] ~[name?] ~[family-name]

~[name]
<2000 names>

~[family-name]
<100 popular family names>

When i run chatette with the above, it takes ages even i only request 1 random example.
Looks like chatette generate all combination, which is (2000^2)*100/2, then random sample one example from it.

Do I understand it right? Could that be, in some way, configured as sampling dynamically. E.e., if I request 1 example, I only randomly generate 1 example?

Thanks!

[SimGus]

Hi,

After checking out the kind of examples you give, there is indeed a problem.

Normally, when asking for a number of examples that is much smaller than the total number of possible examples for the intent, Chatette is configured to only generate the required number of examples. On the other hand, if you request a number of examples that is close to the total number of possible examples, it will generate all possible examples and then take random samples from that list.

That being said, there have been performance issues for a few version (as reported here), which makes me think there certainly is a bug somewhere in the generation code. I have been looking into that but I don't have much time at the moment, so this might still take a while.

Thanks a lot for reporting this!