hyperloglog-sketch-estimation-paper
hyperloglog-sketch-estimation-paper copied to clipboard
oertl
Provide build instructions
I have unsuccessfully attempted to build the paper and a variety of the sub tasks in the build.gradle file. Can you provide dependency and build information so that others can replicate the work here?
There were some missing include directives, which I have already fixed. I have successfully tested all build* gradlew tasks in Ubuntu 18.04 on the Windows Subsystem for Linux. Please do not forget to install the Boost C++ libraries (libboost-dev package) and to copy the source of the dlib C++ library into the folder c++/dlib as mentioned in the c++/dlib/README file. Please, let me know if you experience any further problems.
Thanks. I found that I also needed (building on ubuntu 18.04) to install python3-scipy, python3-matplotlib, dvipng and inkscape to get the figures task to complete successfully.
I have not been able to do ./gradlew paper (pdflatex1 fails since pdflatex exits non-zero), but that is less of a concern since I'm really just after the figures and how that code behaves.
The data generation takes a long time (not at all surprisingly). Do you have an estimate of completion time on your hardware? Edit: From the code it just runs until terminated manually. Is this correct?
Also worth noting for others that default-java on 18.04 brings in java-11 which breaks the build, so don't do that.
The data generation tasks run endlessly and generate HLL examples for different cardinalities. The more examples you have the better the charts will look. The charts presented in the paper are based on 10000 examples whose generation took a couple of days on the used hardware.