I computed both and got different results, The results was not very different though. I imaging it to be the same

asked Aug 20 '13 at 09:02

Ziki's gravatar image

Ziki
1112

edited Aug 20 '13 at 09:05

Did you initialize the weights to the same values? Did you use the same loss and penalty for both models?

(Aug 20 '13 at 09:42) larsmans

One Answer:

There should be no reason why the weights in both should be the same, in fact, the weights between two logistic regressions should not be exactly the same either (this depends a lot on the optimizer you use).

If you optimize the LR exactly with the same training set and the same regularizations, you may get at least the same weight relationships, but not the same numbers.

Also, if you are doing it right, you are doing crossover which means that you add additional randomness to the process and thus you should not expect exactly the same numbers.

answered Aug 20 '13 at 09:43

Leon%20Palafox's gravatar image

Leon Palafox ♦
40857194128

Your answer
toggle preview

powered by OSQA

User submitted content is under Creative Commons: Attribution - Share Alike; Other things copyright (C) 2010, MetaOptimize LLC.