Hello, I plan to run intensive SVM algorithms for a project I am working on—is a MacBook Air sufficient for these intensive computations? I was reading a research paper that used amino acid sequences as the input for an SVM, and they said "the computations were carried out on a Silicon Graphics IRIS Indigo work station (Elan 4000)." I'm really hoping I don't need a fancy machine to run my algorithm. Help?
Pretty sure any modern laptop runs rings around an SGI IRIS Indigo, given that it was released in 1991... ;) Seriously though, you need to elucidate the nature of the problem you're trying to solve, as other commenters have suggested.
can you give more information about what you are trying to achieve? (inbut)
based on the amount of training instance n the complexity of training of non linear SVM between O(n^2) and O(n^3)
unless you're using linear SVM, number of attributes doesn't matter much. If you're using nonlinear SVM, a matrix of size 40,000x40,000 needs to be used, which would require quite a bit of memory, which would require more memory than a typical laptop can handle. There are methods which require less memory at the price of more computation time,
If libsvm fails you, try SVMlight. It may be old, but last time I used it (2 years ago), it did an excellent job, and the man who wrote it also wrote a paper on making large-scale SVM practical (mostly memory-wise)
it's about how the package was implemented, which is why I recommended SVMlight, though you should give libsvm a try since it's better maintained. They probably took memory usage into consideration
I plan to run a binary classifier on a library of 40,000 small molecules. Haven't decided on the number of attributes yet; let's just say 10 attributes for each instance. I'll be running the libsvm package.
Pretty sure any modern laptop runs rings around an SGI IRIS Indigo, given that it was released in 1991... ;) Seriously though, you need to elucidate the nature of the problem you're trying to solve, as other commenters have suggested.
can you give more information about what you are trying to achieve? (inbut) based on the amount of training instance n the complexity of training of non linear SVM between O(n^2) and O(n^3)
What is your classification task? How many samples and how many features? Which package are you running on MacBook.