Closed regislebrun closed 4 years ago
tutorials dependencies can be bigger than the core package, they're meant to be examples
Yes but I was planning to turn the tutorials into tests, waiting for a better solution. The issue with numpy is worth noting as numpy1.18.5 may not be available for some users.
The package networkx is imported in the method plot_with_labels_at_nodes of DimensionTree, which is used only for display purposes and hence should not be necessary in tests.
As for the issue with numpy, I agree that it should be added as a requirement in setup.py. I do not replicate the bug with numpy 1.18.2, I need to find from which version the bug has been fixed.
it works from 1.18.0, 1.17.5 is the latest from 1.17.x and does not work requiring 1.18 is a bit high, as it was only published in december, it could be cool to enable 1.17
The issue with numpy's expand_dims function, due do the use of a list in second argument which was not supported before v1.18.0, will be fixed in the next commit to be compatible with previous versions.
In addition to tensorflow, networkx is also used in some tutorials. I also noted a bug with numpy 1.16.5 & 1.17.5, triggered by
tutorial_tensor_learning_TreeBasedTensorLearning.py
:This issue is solved if I upgrade to numpy 1.18.5. It may be tested in setup.py