[mlpack-git] [mlpack] Class to Finetune deep network (#460)

[stereomatchingkiss]

Provide a class(FineTuneFunction) to finetune the deep network, I write unit test for this class too, the implementation should be correct.

I use MNIST(10000 samples since 60000 take a long time to train), to test the accuracy before finetune and finetune(two sparse autoencoder layers and one softmax layer, the accuracy before finetune is 0.8934, after finetune it raise to 0.9419.

If you think the api are weird, or there are rooms to improve the performance(ex : cache the value), please give some advices. 

You can view, comment on, or merge this pull request online at:

  https://github.com/mlpack/mlpack/pull/460

-- Commit Summary --

  * implement finetune for deep learning
  * 1 : cache probabilities

-- File Changes --

    A src/mlpack/methods/finetune/finetune.hpp (182)
    A src/mlpack/methods/finetune/finetune_impl.hpp (184)
    A src/mlpack/methods/finetune/softmax_finetune.hpp (34)
    M src/mlpack/methods/softmax_regression/softmax_regression_function.cpp (10)
    M src/mlpack/methods/softmax_regression/softmax_regression_function.hpp (9)
    M src/mlpack/tests/CMakeLists.txt (1)
    A src/mlpack/tests/finetune_test.cpp (291)

-- Patch Links --

https://github.com/mlpack/mlpack/pull/460.patch
https://github.com/mlpack/mlpack/pull/460.diff

---
Reply to this email directly or view it on GitHub:
https://github.com/mlpack/mlpack/pull/460
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.cc.gatech.edu/pipermail/mlpack-git/attachments/20151009/15a733b7/attachment.html>