WIP: Update model at runtime

[bakwc]

Ability to add text fragments to a model in runtime, relates to #18

[iosadchiy]

A very useful feature. Is there any chance to see it merged?

Also I don't quite understand how it updates the perfect hash: if the text fragment to be added contains an n-gram that is not present in the hash, how to add it? Don't you have to re-calculate perfect hash from scratch?

[bakwc]

A very useful feature. Is there any chance to see it merged?

I'll try to continue working on this, but not sure when.

Also I don't quite understand how it updates the perfect hash: if the text fragment to be added contains an n-gram that is not present in the hash, how to add it? Don't you have to re-calculate perfect hash from scratch?

I planed to add a separate hash_table for additional words, it would be ok only for adding small amount of new words / sentences at runtime. For example in text editors when you plan to add some exceptions / ignore some errors.

[iosadchiy]

Actually, what I'm trying to do is to first create the model based on n-gram frequencies (#31) and then update it with some domain-specific texts. What could be the best approach here?

It seems like the functionality from this PR is not the best choice (due to the use of an additional hash table).

The other approach is to store all the ngrams in the model (without loading them into memory). This should allow to re-train the model on additional data.

Maybe you'll suggest something better?

[bakwc]

The other approach is to store all the ngrams in the model (without loading them into memory). This should allow to re-train the model on additional data.

It could lead to a performance issues if we will go to disk each time we need to get frequency.

[Jbiloki]

Is there a main function to invoke these commands?

[rprilepskiy]

@bakwc, do you (by any chance) have plans to solve the conflicts?

[bakwc]

sory, currently have no time (

[bakwc]

Done in Pro version.

JamSpellPro is available at jamspell.com