How is the score in v2 calculated?

[kphowell]

The readme describes a score between 0 and 100 that can be used as a threshold to control precision and recall. Can you provide more information on how this score is calculated, please?

Great dataset, by the way. Thank you.

[philipperemy]

@kphowell hey thank you! The scores are calculated based on the frequencies of the names for a given country. For example, the most popular first name in Morocco is Mohamed so Mohamed will have a score of 100.

[KOLANICH]

I don't consier it as solved. I need probabilities, not "scores". How can I get them?

[philipperemy]

@KOLANICH Divide by 100 to have a probability.

[KOLANICH]

@philipperemy, do you really think Martin has probability of 1 to be a surname and just 0.8 to be a name, and 1 + 0.8 = 1.8 > 1? Can I get P("martin" | name) and P("martin" | surname) to use them in naïve Bayes inference (I wanna infer probabilities that a name-surname pair has the right order (first component is considered name, second is considered a surname), that it doesn't have the right order (they are swapped) and that they are not names and surnames at all (i.e. the both components are names or surnames or not in the dataset at all) ).

[philipperemy]

@KOLANICH okay I understand what you want to achieve here.
I dumped the full dataset here (curated of course): #17 (comment).
It contains one CSV per country and on each one <first>,<last>,<gender>,<country>.
You should be able to derive the probabilities you need with this data.

[KOLANICH]

Thanks for the info, in fact I have already seen that issue and that dataset, just didn't want to use it and tried to limit myself with the data already present within the package. BTW, should I send any PRs here or should I leave everything as it is and make an own package?

[philipperemy]

@KOLANICH any PRs is more than welcome! If you think what you will do can be helpful to others, you can open PRs here! Thanks a lot.