[developers] Why no of weights is more than no of unique features in .events file?
chowd at coli.uni-sb.de
Mon Apr 27 23:34:44 CEST 2009
Never mind. I found the reason why those files are not consistent with each
The .event file created inside estimate-model() (and later used by tadm to
create .weights file) is overwritten inside learner-rank-items(). The no. of
items used to write the file inside those two methods are not same. Hence,
the later version of the file contains different no. of unique feature codes
than that of the prior one.
On Mon, Apr 27, 2009 at 12:54 AM, Faisal Chowdhury <chowd at coli.uni-sb.de>wrote:
> I found a mysterious thing in .model.<user>.<pid>.events and .model.<user>.<pid>.weights
> files while running the grid script in LOGON. The no. of weights in .weights
> file is bigger than the highest feature code inside the .events file. I did
> some experiments of TADM with some dummy data and found that the no. weights
> predicted by TADM is equal to the no. of unique features indexes, i.e., the
> no. of weights should be the highest feature code + 1. Could you please
> explian why .events and .weights files are not consistent on this?
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the developers