Something seems wrong with bayes classify method about classifier-reborn HOT 5 CLOSED

@marciovicente Sorry for the delayed reply. I'll see if I can reproduce it. What does the structure of @training look like?

from classifier-reborn.

Basically the training data looks like:

@training = [
    ['text in portuguese', :t],
    ['another text in portuguese', :t],
    ['one more', :f],
    # [...]
    ['this is in english', :m]
]

from classifier-reborn.

Is it possible to upload the dataset being used as well?

I'm able to replicate a scenario with a toy dataset, where I also get odd outcomes:

{"T"=>418, "F"=>182, "M"=>1}

...But that happens when the T, F, and M sentences are similar to each other, making it easy for false positives and false negatives to occur (computers can make mistakes too). Since Polish and French are both "romance" languages, it's very easy to imagine overlap between the two that might cause the machine to make a classification error. But I can't be sure (as I don't have the actual dataset and I don't really know much about Bayesianism other than the fact that it relies on probabilities).

from classifier-reborn.

@marciovicente If you can share more data, perhaps I can help.

from classifier-reborn.

closing due to inactivity

from classifier-reborn.