Error installing from neurips2019 branch
bengreenberg5 opened this issue · 1 comments
Hello, I have had some difficulty setting up the environment to run experiments in the neurips2019 branch. (I have been able to use the master code to train/evaluate models, but wanted to run the experiment notebooks as well.)
I cloned the neurips2019 branch, created a new conda environment with python=3.7, ran install.sh, and installed tensorflow==1.13.1 and mpi4py manually as specified in the instructions. When running run_tests.py, I encountered errors due to missing the smmap and pytz libraries; after installing these, I hit another error: ValueError: numpy.ufunc size changed, may indicate binary incompatibility. Expected 216 from C header, got 192 from PyObject This happens because of an issue with the pandas install, which (I think) results from incompatibility with the specified numpy version.
Could someone help me get past these installation problems?
Sorry for the late reply – I'm currently very strapped for time but should be able to look into this in a couple of weeks. I hope that's not too late? I'm guessing that you're correct in that it's a version issue. I would maybe try to install both numpy and pandas with conda too, just as to minimize the chance of incompatibilities between them?