dice algorithm NaN

[Globegitter]

Just trying out this algorithm and seems the dice algorithm has some minor bugs (or I am not understanding it quite right):

These are the results I am getting with strings of length 0 and length 1, could this have anything to do with the input being characters rather than actual strings? Is that as intended?

[Yomguithereal]

Hello @Globegitter, Well this is rather unfortunate. I'll give it a look as soon as possible. This bugs applies to the 0.1.9 version or the 0.1.8?

For other cases, does the algorithm work correctly?

[Globegitter]

@Yomguithereal It applies to both. Otherwise it seems to be working really well - thanks for the library, it is really useful.

[Yomguithereal]

@Globegitter, I've checked this and can affirm the bug comes from the clojure part and therefore replicates into its JavaScript counterpart.

I can fix it but I have a problem here and you might be able to help me:

The Dice coefficient works using bigrams. So, traditionnally, if you compare h and h, this will return 0, which is a total nonsense since both strings are the same.

So here is the choice I have to make:

• Follow most of the classical mathematical implementations of the algorithm and bear that on strings with less than two characters the coefficient will produce nonsense.
• Create a finer implementation that would deal with this edge case and return correct similarities on an intellectual level.

Any opinion?

[Yomguithereal]

I've fixed the implementation. You can install the latest dev version with the following command for node if needed:

npm i git+https://github.com/Yomguithereal/clj-fuzzy.git

[Globegitter]

Oh that is great thank you! How did you resolve it then?

[Yomguithereal]

Second choice. I found other libraries - in python notably - that prefer to fix the rationale of the algorithm. So I went with that so now h / h --> 1.0.

[Globegitter]

Awesome thank you, will test asap.