This fixes issue #8 . The cause is we build our library against the tensorflow hosted by conda. It turns out that conda's tensorflow uses new CXX11 ABI while the official tensorflow (accessible using pip install) uses old ABI, which causes the error. This PR switches the tensorflow from conda's to the official one because tfdlpack currently does not support conda install.
This fixes issue #8 . The cause is we build our library against the tensorflow hosted by conda. It turns out that conda's tensorflow uses new CXX11 ABI while the official tensorflow (accessible using pip install) uses old ABI, which causes the error. This PR switches the tensorflow from conda's to the official one because tfdlpack currently does not support conda install.
The PR also enables tfdlpack on python 3.5.