Hi Tim, Thanks! I'm actually looking forward to see a version update of your great HDF5.jl. And BTW: I have been thinking about data randomization. How inefficient do you think it will be if I read hdf5_dset[:,:,:,i] for i be 100 random numbers within the index range, comparing to reading hdf5_dset[:,:,:,k+1:k+100], reading 100 consecutive examples (no randomization here) all at one time? Is there a recommended / better way of doing random access in HDF5 (HDF5.jl)? Thank you very much!
Best, Chiyuan On Friday, November 28, 2014 2:51:29 PM UTC-5, Tim Holy wrote: > > Cool stuff! > > --Tim > > On Friday, November 28, 2014 07:42:47 AM Chiyuan Zhang wrote: > > Hi all, > > > > Mocha.jl <https://github.com/pluskid/Mocha.jl> is a Deep Learning > framework > > for Julia <http://julialang.org/>, inspired by the C++ Deep Learning > > framework Caffe <http://caffe.berkeleyvision.org/>. > > > > Please checkout the new IJulia Notebook demo of using pre-trained CNN on > > imagenet to do image classification: > > > http://nbviewer.ipython.org/github/pluskid/Mocha.jl/blob/master/examples/iju > > lia/ilsvrc12/imagenet-classifier.ipynb > > > > Here are detailed change log since the last release: > > > > v0.0.3 2014.11.27 > > > > - Interface > > - IJulia-notebook example > > - Image classifier wrapper > > - Network > > - Data transformers for data layers > > - Argmax, Crop, Reshape, HDF5 Output, Weighted Softmax-loss Layers > > - Infrastructure > > - Unit tests are extended to cover all layers in both Float32 and > > Float64 > > - Compatibility with Julia v0.3.3 and v0.4 nightly build > > - Documentation > > - Complete User's Guide > > - Tutorial on image classification with pre-trained imagenet model > > > > > > Best, > > pluskid > >