Hi all, Below is a progress report on the Deeplearning integration
The previous two weeks I mainly dedicated for testing what is already implemented for errors and potential enhancements. *Getting the deeplearning component to work* After rigorously testing the deeplearning component, it seems to be working well with all the datasets tested so far. There were few bugs in the implementation (i.e. onExBarrierCompletion error, conversion to enum) which have been resolved with the help of WSO2 team and H-2-O team. *Testing the deeplearning component* So far DL component has been tested against 4 datasets. They are, *Iris, MNIST, Leaf, Action Recognition* Deeplearning has shown remarkable performance with all the datasets except for the leaf dataset. (The performance for the leaf dataset does not exceed the Linear Reg - L-BFGS by a lot) I speculate this could be due to a linearity in data (so the model is too complicated). For all the other datasets, DL reaches 95%+ without much effort *Fixing minor bugs and Enhancements* Few minor bugs were fixed such as, Input sanitization error for Stacked Autoencoders Code formatting and changes suggested in the review Enhancements include, Introducing activation function selection -- Regards, Thushan Ganegedara School of IT University of Sydney, Australia
_______________________________________________ Dev mailing list Dev@wso2.org http://wso2.org/cgi-bin/mailman/listinfo/dev
