I'm coming to Julia from years of work in Matlab, but had a hard time figuring out how to use Julia to access the statistics.hdf5 file that was recommended for the solver in the mnist.rst example that shows up in the docs. Note of clarity on this...the source file provided in the examples folder had the file type as statistics.jld but I wrote my follow-along script from the example in the docs that says to use a .hdf5 format statistics file. The first commit in this pull request changes the mnist.rst file in the docs to recommend a .jld file which I found MUCH easier to use.
After getting access to the stats file I thought I might improve the tutorial a little bit for folks new to machine learning by walking them through the process of generating a learning curve from the stats file with rationale for why it is important and some background on the noise they will see with SGD.
I'm coming to Julia from years of work in Matlab, but had a hard time figuring out how to use Julia to access the
statistics.hdf5file that was recommended for the solver in themnist.rstexample that shows up in the docs. Note of clarity on this...the source file provided in the examples folder had the file type asstatistics.jldbut I wrote my follow-along script from the example in the docs that says to use a.hdf5format statistics file. The first commit in this pull request changes themnist.rstfile in the docs to recommend a.jldfile which I found MUCH easier to use.After getting access to the stats file I thought I might improve the tutorial a little bit for folks new to machine learning by walking them through the process of generating a learning curve from the stats file with rationale for why it is important and some background on the noise they will see with SGD.
Hope this helps!