search

peterwittek / somoclu

Massively parallel self-organizing maps: accelerate training on multicore CPUs, GPUs, and clusters
https://peterwittek.github.io/somoclu/
MIT License
266 stars 69 forks source link

MATLAB interfece Batch algorithm #153

Closed bedassa closed 4 years ago

bedassa commented 4 years ago

I understood that somoclu is parallel and batch training algorithm. I run the test somoclu_train in the MATLAB interfece and sTrain shows a seq algorithm. I need batch algorithm as I am dealing with large dataset. Please, I need a tip how to change it into batch algorithm

xgdgsc commented 4 years ago

What do you mean by not batch? The cpp implementation is batch.

bedassa commented 4 years ago

I run mex_interface_test.m which is the test file in the somoclu/src/Matlab . The sTrain indicates seq algorithm. I need how to change to batch. Thank you type: 'som_train' algorithm: 'seq' data_name: 'D' neigh: 'gaussian' mask: [3×1 double] radius_ini: 50 radius_fin: 1 alpha_ini: NaN alpha_type: 'linear' trainlen: 100 time: '24-Jul-2020 16:06:18' radius_cooling: 'linear' kernel_type: 0 compact_support: 0 neighborhood: 'gaussian' scale0: 0.1000 scaleN: 0.0100 stdCoeff: 0.5000 verbose: 0 gaussian: 1

xgdgsc commented 4 years ago

That' s only a string filling to reuse som toolbox data structure. You can change https://github.com/peterwittek/somoclu/blob/master/src/MATLAB/somoclu_train.m#L72 if you want.(And welcome a PR!) The call to somoclu cpp code is calling a batch algorithm.

bedassa commented 4 years ago

That is awesome. Thanks!!