|
Recently came upon the post on Practical Machine Learning Tricks. It was a real eye-opener regarding the some basic tips that a machine learner needs to use. Moreover, it tells that much of these tricks of the trade are left to folk wisdom. So I wanted to get some pointers to resources(like tutorials/blog posts) that talk about the engineering aspects of machine learning. I am looking to carry the conversation on engineering tricks/methods like the ones mentioned in the link. |
|
Just check out all of Alexandre's and Olivier's answers to the "why doesn't my algorithm work" questions here on MO. Also: use the source, luke! (try scikit-learn)
This answer is marked "community wiki".
thanks for the answer. And yes i am using scikit-learn :). I was interested in finding if some similar resources were documented elsewhere so that it will be useful. will go through the answers...
(Aug 21 '12 at 09:06)
satyadileep
By "use the source", I think that Andy meant: read the source code.
(Aug 29 '12 at 02:58)
Gael Varoquaux
|
|
Efficient Backprop (LeCun et al) has a lot of good advice about training neural nets. Yoshua Bengio has a new paper on how to train deep networks, "Practical recommendations for gradient-based training of deep architectures". You're right that it's regrettable that people don't talk about these details in publications as often as they should. It's one of the reasons I started this forum. Efficient Backprop is a really excellent reference for supervised neural nets.
(Aug 24 '12 at 09:50)
ogrisel
|
