Hi all,
I am writing a piece of code in the above mentioned files to get the best path scores. The pseudo codes are as follows.
Token tok= result.getBestToken();
double score=(double)tok.getScore();
Now when I print this score it is a negative number (since it is in log form) and the values have to be less than zero. But I get one value which is fairly above zero. Which is implying that linear score is much much greater than one. Well my concept of the score is that this is best path's overall probability sum (since it is log). It would be a product in the linear domain. And should remain less than or equal to one.
Or is it that I am looking at things in the wrong direction?
Regards
Adil Raja
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Not necessarily less than 1.... Think of a single one-dimensional gaussian, to simplify matters. Mean m, variance 0.00001. When you compute the value at a given point, say, m, the output you get is >> 1. When you integrate over all possible values, then the sum is one. In this case, for each frame, you're computing the score at a point, and then multiplying the values you get (or summing the log-values).
--Evandro
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi all,
I am writing a piece of code in the above mentioned files to get the best path scores. The pseudo codes are as follows.
Token tok= result.getBestToken();
double score=(double)tok.getScore();
Now when I print this score it is a negative number (since it is in log form) and the values have to be less than zero. But I get one value which is fairly above zero. Which is implying that linear score is much much greater than one. Well my concept of the score is that this is best path's overall probability sum (since it is log). It would be a product in the linear domain. And should remain less than or equal to one.
Or is it that I am looking at things in the wrong direction?
Regards
Adil Raja
Adil,
Not necessarily less than 1.... Think of a single one-dimensional gaussian, to simplify matters. Mean m, variance 0.00001. When you compute the value at a given point, say, m, the output you get is >> 1. When you integrate over all possible values, then the sum is one. In this case, for each frame, you're computing the score at a point, and then multiplying the values you get (or summing the log-values).
--Evandro