Everybody's Got Problems
Just when you think that the blogosphere consists almost entirely of cat diaries, teenage angst-fests, and uninformed political opinion-mongering, you come across a blog with posts such as this:
I looked further into the nondeterministic behavior of the forward selection algorithm, and I found out that it is not the algorithm at all. Nor is it a product of recent changes I've made in my modules, as I checked it with older versions. It seems to be a result of something in D2K (my modules, or Weka, or some combination) that, when I evaluate just one feature vector (a population size of 1 for a GA), the result is always the same. When I evalute 2 or more in one population, the results differ. I can't figure out the reason for this. The problem does not happen if I restart D2K between every evaluation; that is, if I evaluate vectors X and Y, then restart, then evaluate X and Y again, the results will always be identical. This is not the case if I evaluate X and Y, then run the itinerary agan without restarting D2K. The results are always similar, so I don't think that it indicates the classifier isn't working per se; it just indicates that there is some random process going on that I can't fix to happen the same way every time. I'm not going to try to fix this right now, given that there is no obvious solution and this doesn't seem to be a big problem. ...
Rebecca's Research Blog is where Rebecca Fiebrink of Montreal keeps progress notes on her thesis project. Good luck, Rebecca! That D2K can be a real bitch.
Technorati Tags:
blogosphere, blogging, blogs, university, universities, college, theses, research, math, statistics, mathematics, writing, Montreal
I looked further into the nondeterministic behavior of the forward selection algorithm, and I found out that it is not the algorithm at all. Nor is it a product of recent changes I've made in my modules, as I checked it with older versions. It seems to be a result of something in D2K (my modules, or Weka, or some combination) that, when I evaluate just one feature vector (a population size of 1 for a GA), the result is always the same. When I evalute 2 or more in one population, the results differ. I can't figure out the reason for this. The problem does not happen if I restart D2K between every evaluation; that is, if I evaluate vectors X and Y, then restart, then evaluate X and Y again, the results will always be identical. This is not the case if I evaluate X and Y, then run the itinerary agan without restarting D2K. The results are always similar, so I don't think that it indicates the classifier isn't working per se; it just indicates that there is some random process going on that I can't fix to happen the same way every time. I'm not going to try to fix this right now, given that there is no obvious solution and this doesn't seem to be a big problem. ...
Rebecca's Research Blog is where Rebecca Fiebrink of Montreal keeps progress notes on her thesis project. Good luck, Rebecca! That D2K can be a real bitch.
Technorati Tags:
blogosphere, blogging, blogs, university, universities, college, theses, research, math, statistics, mathematics, writing, Montreal
1 Comments:
awesome, well done.
the internet(s) are a strange place, atren't they. I tell you, the truth is out there somewhere. but it has been equally censored and diluted.
Post a Comment
Links to this post:
Create a Link
<< Home