accord-net / framework

Machine learning, computer vision, statistics and general scientific computing for .NET
http://accord-framework.net
GNU Lesser General Public License v2.1
4.49k stars 1.99k forks source link

Principal Component Analysis is so slow #266

Open MoamenMohamed opened 8 years ago

MoamenMohamed commented 8 years ago

i have a dataset with dimensions of 900x1400 it takes over 1 minute to compute the components but in a tool like matlab it takes few seconds. Also when the data get bigger 5000x1400 the time problem is much worse (10 minutes at least). i don't know if i am doing something wrong or what.

cesarsouza commented 8 years ago

Hi Moamen!

Please, can you let me know how many samples you have in your dataset? There has been some speed improvements in the latest release regarding PCA. I will (or if someone else would like to, please try it to) grab a dataset that is representative of this problem and check the performance under a profiler. Also please make sure to not use any Obsolete/Deprecated methods, as they will for sure run slower..

Currently the framework is not using vectorized operations as it currently doesn't have dependencies on BLAS or Intel MKL. It is possible that this can change in the future (Issue #277).