We agree it'd be great to have a fix here, since a number of us have hit it, but we're not sure what the right solution is presently.

When iterating through token counts in the loop at line 553 (https://github.com/allenai/allennlp/blob/master/allennlp/data/vocabulary.py#L553), the get method is used to either get a min_count for that namespace, or 1 if that namespace isn't in min_count. One way to address this issue would be to have a try block that tries to access min_count[namespace]. If the namespace doesn't exist, a KeyError occurs, and a ConfigurationError could be raised.

An alternate method is iterating through all the keys in min_count before the loop at line 553, and checking if that key is in counter. If it isn't, a ConfigurationError is raised.

If you think either solution is suitable, I could implement it.

It seems to me like the second solution is correct - (I think) we want to fall back to a default of 1 for namespaces that aren't mentioned, so the first solution doesn't work. @nelson-liu, what do you think?

@nelson-liu @matt-gardner If this issue is still pending, I would like to take this up. If so, should I go forward with the second approach to solve it?

Yes, @nadgeri14, that would be great. The line numbers referenced above have changed, so the place to do it is probably right above this line, at the top of _extend:

@matt-gardner Thanks, that was really helpful.

I need a small help: Could you please provide a test case, as it will help me to validate my solution before submitting the PR.

Thanks in advance for the help.

You just need a counter with some set of keys, and a min_count parameter that has keys not present in the counter.