0

I've just implemented a naive Bayesian classifier and found out about the Fisher method (Linear discriminant analysis and Bayes rule: classification) while looking for ways to improve it. I'm very new to this field.

My raw data model is like the following (for sentiment analysis):

{"I like the movie a lot", "positive", 1000}
{"I hate the movie a lot", "negative", 100}
...

As you see, I have only two classes, possibly more, and the third value is weight values to count when getting probabilities.

So when given this kind of data, and to get the PDF like below,

$$PDF(x|k) = \frac {e^{-d/2}} {(2\pi)^{p/2}\sqrt{|S|})}$$

I have no idea how to approach this. What should I set as discriminants? Where should I start to matrix-ify my data to get the covariance? Once I model the data, the next step seems to be relatively easier with just calculation.

In short, what should I do to get the values of $p$ discriminants from the data {"I like the movie a lot", "positive", 1000}, with the first value as training text, send as class value, third as weight value.

Community
  • 1
EPSILONsdfsdfdsf
  • 101
  • 4
  • It is unclear if you want to implement Discriminant analysis or Bayes classifier. DA first extracts the discriminants. Then it classifies (in a manner of a Bayes classifier) using _those_. If you need DA you ought to read more pages (including this site) about it, to stop being so `new to this field`. – ttnphns Oct 19 '13 at 07:11
  • I see, thanks, I was not sure how to convert sentences into numbers... – EPSILONsdfsdfdsf Oct 19 '13 at 19:45

0 Answers0