search

Xtra-Computing / thundersvm

ThunderSVM: A Fast SVM Library on GPUs and CPUs
Apache License 2.0
1.56k stars 217 forks source link

Working with MATLAB datasets stored as variables in the workspace #193

Open jl1101 opened 4 years ago

jl1101 commented 4 years ago

Hi,

I can read from the example that when implemented on MATLAB, the svm_train_matlab and svm_predict_matlab commands have to refer to a dataset that is stored on the harddrive as a txt file. Is it possible for thundersvm to work directly with datasets stored as variables in MATLAB workspace the way LIBSVM does? Exporting the datasets as txt files are not very plausible for me because my model employs cross-validation and random sampling: storing a new txt file into the harddrive in each iteration (tens of thousands) is not efficient computationally (especially when the dataset is very large).

Could any one please kindly suggest?

Thank you.

zeyiwen commented 4 years ago

The Matlab interface currently doesn't support the functionality. You are more than welcome to contribute.