Jason (jcreed) wrote,
Jason
jcreed

-

Finally found out what was wrong with the code at work.
There was a special case for the event that the feature
count for some feature was zero, which set the gradient
to zero. Removing this hack made things behave how I expected.

Also my old Model Theory book was at work after all.

New terse LF spec from kw.

Hacked on the conlang thing some more.

Listened to disintegrationphactor again.
I had forgotten how much I liked how `social engineering' turned out.
Subscribe
  • Post a new comment

    Error

    Anonymous comments are disabled in this journal

    default userpic

    Your reply will be screened

    Your IP address will be recorded 

  • 0 comments