vatlab / sos-julia

SoS extension for Julia
BSD 3-Clause "New" or "Revised" License
0 stars 1 forks source link

Fail to get dataframe with pandas 0.24 #19

Open BoPeng opened 4 years ago

BoPeng commented 4 years ago

Travis CI fails with message

 ValueError: Cell produces error message: Failed to get variable: module 'pandas' has no attribute 'lib'. Use expect_error=True to suppress this error if needed.

which seems to be related to an incompatible change of pandas. see https://github.com/Neurosim-lab/netpyne/issues/406 for details.

BoPeng commented 4 years ago

Tried to use latter version of feature-format (https://github.com/vatlab/sos-julia/commit/623d7be9e6c49e48c2581092c9fc7bfc5c30a717) and see that feather will not load.

.inpavis/miniconda/envs/python_test/lib/python3.6/site-packages/pyarrow/lib.cpython-36m-x86_64-linux-gnu.so: undefined symbol: _ZNK5arrow5Field8ToStringB5cxx11Ev

I guess this is why version 0.3.1 was specified.

@mathieuboudreau Are you on linux? What version of feather-format do you have?

feather-format 0.4.1 works fine on my mac here.

I knew it would need a lot of work to get travis running so I released sos-julia 0.18.3 after the tests passed locally.

mathieuboudreau commented 4 years ago

@BoPeng currently I'm testing locally on my Macbook, but we also use Linux through Docker builds on MyBinder. I haven't run the tests on my local computer though, only through Travis in a fork.

BoPeng commented 4 years ago

Everything is fine on my mac so I am firing up a CentOS VM now, which should be closer to Travis.

BoPeng commented 4 years ago

With the latest addition of using Pkg, %get a where a=[ 1 2] works on CentOS + JupyterLab without the need to pre-install DataFrames.

The travis problem might be with the old version of feather-format, but the new version of feather-format does not work there.