pylint | It's not just a linter that annoys you! | Code Analyzer library

I upgraded pylint to a more recent version.

A solution for this is being actively discussed and developed at the Pylint project. The workarounds until a fix is merged are either to use an earlier version of VS Code (before January 2022) or Pylint (below 2.12.2). If the latter is desired, you can download a local copy and specify a custom path to Pylint in the Python extension settings.

Edited answer following additional experimentation

I'm not a python expert, but after a couple more experiments I think I get the information that you need

In my first experiment, I used several modules that were not in a package. It appeared when using different ways to do the imports that pyreverse shows only the classes of the modules that are mentioned on the command line.

While I initially assumed that the issue was related to a stricter import syntax, it turned out that in reality it just worked as designed and documented in the man page: pyreverse shows in the diagram only the classes of the modules listed in the pyreverse command line.

So with a little project with, using almost your definitions in files main.py,a.py and b.py the easy workaround to get all the classes in the diagram was to use pyresverse main.py a.py b.py on a single command line. It generates the diagram:

It appears however that the result also depends on the PYTHONPATH, the current directory from where you call pyreverse and the path used to specify the module, as these elements can influence the finding of the right import.

I renamed then main.py into __init__.py to make a package. I can then use pyreverse providing only the directory name of the package (or . if it's the current working directory). Then all the files of the packages are processed for making the class diagram, withoout a need to enumerate them.

Using pyreverse on the package from within the package directory works with your import syntax and produced a similar diagram as above. However, running pyreverse from its parent directory produces 2 classes side by side without relationship. This means that we are dealing with two different classes B one of which was not found. Using the relative import solved the issue.

I found it by the way useful to add option -m y to disambiguate the class names by adding the module name:

I could verify experimentally: whenever I get unlinked classes, launching the module in python from the same current working directory as pyreverse caused an import error.

Wrapping the transform_targets_for_output method with a try-except-raise clause and applying tf.data.experimental.ignore_errors to the dataset seems to actually work:

This seems to be a bug in Tensorflow 2.7 when using model.save combined with the parameter save_format="tf", which is set by default. The layers RandomFlip, RandomRotation, RandomZoom, and RandomContrast are causing the problems, since they are not serializable. Interestingly, the Rescaling layer can be saved without any problems. A workaround would be to simply save your model with the older Keras H5 format model.save("test", save_format='h5'):

Using sed:

Do not use pylint-fail-under, pylint has a fail-under option since pylint 2.5.0, and it's maintener will not update this package for newer pylint.

Change pylint-fail-under --fail_under 8.0 to pylint --fail-under=8.0 and remove the dependency to pylint-fail-under.

See also https://github.com/PyCQA/pylint/issues/5405, and: https://github.com/TNThieding/pylint-fail-under/issues/8#issuecomment-626369567

Here is a complete running example based on your dataset running in Google Colab:

Need to down-grade to Keras 1.0.2 If works then great, otherwise I will tell other solution.