is muliple classification possible?

[GoogleCodeExporter]

First of all, i want to thank the author for this implementation. It brings a 
lot of useful informations, it is often easy to understand and use but can i 
perform multiple classification with this code? 

Original issue reported on code.google.com by alain.ti...@gmail.com on 11 Dec 2012 at 7:19

[GoogleCodeExporter]

yup, RFs can do what decision trees can do including inherent support for 
multi-class classification. I guess you meant multi-class classification right?

Original comment by abhirana on 11 Dec 2012 at 10:49

• Changed state: Accepted

[GoogleCodeExporter]

Thank you for your fast answer Sir and yes i mean multi-class support. i have 
looked to the example provided with the implementation (twonorm) and there are 
only tow labels used -1 and +1. So i thought that the only available 
classification is the binary one. now i understand that i can use 1, 2, 3,... 
like labels (am i right?). Thank you very much.

Original comment by alain.ti...@gmail.com on 12 Dec 2012 at 9:30

[GoogleCodeExporter]

Yup,  you can use any numerical labels you want,  that is any integer label

Original comment by abhirana on 12 Dec 2012 at 9:48

[GoogleCodeExporter]

[deleted comment]

[GoogleCodeExporter]

[deleted comment]

[GoogleCodeExporter]

[deleted comment]

[GoogleCodeExporter]

Hi, i am using a 3804729*60 double matrix (841705 ko) for training and when i 
run the program, it produces an out of memory error within the classRF_train 
function (line 347). however i have a 64 bits machine with 8 go Ram. when i 
tape memory in matlab it returns:

Maximum possible array:              12026 MB (1.261e+10 bytes) *
Memory available for all arrays:     12026 MB (1.261e+10 bytes) *
Memory used by MATLAB:                2345 MB (2.459e+09 bytes)
Physical Memory (RAM):                8174 MB (8.571e+09 bytes)

is there a limit for the input matrix sir? i am using only 100 trees

Original comment by alain.ti...@gmail.com on 13 Dec 2012 at 12:50

[GoogleCodeExporter]

you will require more memory for internal bookkeeping. roughly 6 x N x Ntree. 
Anyways i think you may also run into computational issue as RF may not scale 
to that many examples and give you results in a reasonable time.

Original comment by abhirana on 13 Dec 2012 at 9:22

[GoogleCodeExporter]

i have avoided the memory error problem by increasing my virtual memory. The 
training process took 6 hours and i saved the model like a matlab variable 
using this cmd:
save -v7.3 LModel.mat model
but it results in a very big file ~=5 Go. is this normal?

Original comment by alain.ti...@gmail.com on 14 Dec 2012 at 1:41

[GoogleCodeExporter]

yup. this is normal. 

right now the variables are stored in such a way that although they are mainly 
zeros they are not compressed. you could maybe do a sparse for the larger 
variables and then when you load them back unsparse them.

Original comment by abhirana on 18 Dec 2012 at 4:58