|
Hi all, I am new to machine learning stuff. I have some theoretical knowledge like classification, regression etc, neural networks and blah blah. I am plannin to model a multiclass classifier. But, before thinking about implementing, the first question arised, how to represent the features while making computation. I didn't get any idea about it.I want really understand how does classifier classify and how they are implemented. Any example,like taking a syntactically parse tree and taking different features like distance between two parse constituent would be higly appreciated. Any useful resource for making concept clear will also be higly appreciated. |
|
Numbers. Features are usually represented as numbers, but there are several types. As a very basic introduction:
That was the intro, there is much more information here, which also contains the "correct" names as used in statistics (and hence, machine learning). Oh, and as for programming-based representation, you usually store them in a two dimensional array X such that X[i][j] is the value of the jth feature for the ith instance. (Some people put those indices around the other way, which makes things faster for some computation and slower for others. I prefer [i][j].) @robert, thank you very much for the clear answer. In addition to it, I would like to add some additional question here, In neural network, I have some feeding points. Why there are different feeding point for different features. Why not single feature vector representing all the features?And one more thing is,how do you train the date? do you first get all the features first or you get features for each of the training sample and evaluate it?
(Feb 09 '12 at 17:25)
Kuri_kuri
It sounds like you have a non-"basic" neural network. It's not my area though, so you would be better off creating a new question (don't forget to say which neural network you are using!).
(Feb 09 '12 at 22:56)
Robert Layton
|
