Bug in collecting counts?

[dgel]

shouldn't line 225 use j+1 instead of j?

emp_feat += DiagonalAlignment::Feature(j, i, trg.size(), src.size()) * p;

in the surrounding loops, i ranges from 1 to src.size() inclusive and j from 0 to trg.size() exclusive. Further up in the code, j+1 is used as well, and I think the feature calculation assumes the same.

This bug would bias it towards a flatter distribution I think, but I'm not sure.

[redpony]

You either pay for this with a slightly off-center prior either when j=0 or
j=trg.size() - 1-- I noticed this after I had run all the experiments and
decided not to fix it, but a couple people have pointed it out, so it's
probably worth fixing at some point...

On Mon, Sep 23, 2013 at 12:27 PM, dreugeworst [email protected]:

shouldn't line 225 use j+1 instead of j?

emp_feat += DiagonalAlignment::Feature(j, i, trg.size(), src.size()) * p;

in the surrounding loops, i ranges from 1 to src.size() inclusive and j
from 0 to trg.size() exclusive. Further up in the code, j+1 is used as
well, and I think the feature calculation assumes the same.

This bug would bias it towards a flatter distribution I think, but I'm not
sure.

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/4
.

[dgel]

You either pay for this with a slightly off-center prior either when j=0 or j=trg.size() - 1

I'm not sure I quite follow this. Right now when collecting counts the features used are
calculated using 0-based indexing, while the rest of the program uses 1-based indexing.
The diagonal is calculated using the same (1-based). When using j+1 in that bit of code, doesn't it
just fix it to do what the paper says? How does that put the prior off-center?