Shark-ML/Shark

Data_Download acting up again

Closed this issue · 11 comments

Since a few days, afte ra longer mldata downtime (i think it still down):

101/145 Test #101: Data_Download ...........................................***Failed 152.63 sec
Running 3 test cases...
/home/travis/build/Shark-ML/Shark/Test/Data/Download.cpp(40): error in "Data_Download_URL": Got exception [download] http protocol violation
/home/travis/build/Shark-ML/Shark/Test/Data/Download.cpp(63): error in "Data_Download_MLData": Got exception [download] http protocol violation
*** 2 failures detected in test suite "ML_Download"

how good is a binding to a webpage that is always down?

TGlas commented

mldata.org does not seem to be well-maintained already for years. The simplest solution is to link to our self-hosted data used in the "hello world" tutorial at http://www.shark-ml.org/data/quickstart-train.csv.

TGlas commented

Let remove mldata.

TGlas commented

A different issue is that we only support http, not https. This will be an increasing issue in the future. That would cost us an openssl dependency, which could be optional. Would you consider that worth the hassle?

TGlas commented

If we remove mldata.org support, then we could also remove support for the HDF5 data set format. As far as I know, mldata is the only platform that pushed the format, and as far as I can tell it was not widely adopted.

TGlas commented

mldata support removed, docs, tests and examples updated accordingly

TGlas commented

Just to be sure: "Fully agreed. Kill it." refers to HDF5 support, not to my short post above?

TGlas commented

done.