DataLayer to provide data from LIBSVM format file

[tnarihi]

I've just implemented a new layer which provides data from LIBSVM format file.

<label> <index1>:<feature1> ... <index M>:<feature M>

Here is an example of using this layer: http://nbviewer.ipython.org/github/tnarihi/caffe/blob/libsvmdatalayer/examples/libsvm_format_data_into_caffe.ipynb

[sguada]

Adding more data layers is not a sustainable approach. Please take a look at #1568 and try to simplify your approach.

[shelhamer]

@tnarihi thanks for adding another data format, but as @sguada mentioned we are trying to come up with a more general, less redundant data pipeline. At the same time we are working on making it easier to have "personal" layers that will build with Caffe but don't need to be merged in the main project.

Your libsvm layer could be one of these contributed layers.

[tnarihi]

@sguada @shelhamer Thanks for the comments ! More general data layer sounds good. I will take a look at the PRs. The "personal" layer concept is also interesting. Is there any PR regarding it?

[shelhamer]

@tnarihi this could be done by a Python data layer #1703 or through de-centralized / personal layers as discussed in #1896 with a proof-of-concept in #2036. Closing to avoid adding more data layers since the data pipeline needs further work and special formats can be handled by Python. Thanks for sharing your code!