winkjs / wink-naive-bayes-text-classifier Goto Github PK

I got unknown prediction from this set:
nbc.learn( 'hi', 'greeting' ); nbc.learn( 'hello', 'greeting' ); nbc.learn( 'hey', 'greeting' ); nbc.learn( 'nice to meet you', 'greeting' );

and:
console.log( nbc.computeOdds( 'hii' ) );

is It ok?

Hey everyone!

So, I couldn't find any docs about this so I was making my own take on this, by making a history, which bot goes thru when question is asked and well thinks of answers to them too. but I got a problem. Wink uses negative numbers too! so im wondering if I can either: Use 0 to 1 numbers or if there is some way around it too!

Last time I asked you guys, you were REALLY helpful! Thanks for the work and stuff you make is amazing!

(also rephrasing what I mean by open questions: basically when asking bot lets say "who was nikola tesla", and then asking "did he love pigeons" I want it to know that I mean nikola tesla and do the processing then 😊)

Does anyone do any math conversions on computeOdds that they could share? I'm looking for output values between 0 and 1, for using the data in charts and I'm not sure how the base-2 odds would work?

Like express (HTTP server), and unlike other wink libraries, NBC needs to be instantiated before using.

Express

var express = require('express');
var app = express();

NBC

var nbc = require( 'wink-naive-bayes-text-classifier' )();

In our source code, could we:

module.exports = textNBC();

The function doesn't have any arguments so there aren't two ways of doing this any way.

So yeah title says it all, I have been playing around with wink js and I just basically made a small dataset, which works all fine but on ONE data. so I got like basics right "hello" and "good bye" "how are you" and I have one that is a type of function "what time is it" again, I was playing around so all of em work but "how are you" it says "Unknown".

^^ thats the how are you message.
and here are other ones:



so yeah. can't figure out the reason it just says unkown idk...

In an app I've written I categorise downloaded banking transactions using manually created rules. I thought ML could be a handy alternative for this...

I have been trying Wink for several days, but haven't been really successful. I always find approx 10% false categories being assigned and only a few 'unknown'. 'Unknown' is acceptable because these can be assigned manually, but false assignments are detrimental when analysing income and spending.

Is Bayes really suitable for this type of classifications?

A possible issue I can think of, is that transactions of a single payee can be assigned to different categories. Eg. a warehouse store has multiple departments and hence multiple categories (clothing, personal health, entertainment...). The store shows the same name in the transactions, but is assigned to different categories.

• So I created a word list of all words that occur in payees with multiple categories. Still too many errors...
• I created a list of payees with multiple categories and leave them out of the learning dataset. No success either...

Any thoughts/suggestions on this?

Is there a way to currently do this or does it not yet exist?