thunderhoser/ai2es_xai_course

Notebooks for AI2ES (NSF Institute for Research on Trustworthy Artificial Intelligence in Weather, Climate, and Coastal Oceanography) short course on XAI (explainable artificial intelligence).

ai2es_xai_course's Issues

lecture 3 notebook: error running cell in Google Colab

I am trying to run this notebook with no changes: https://colab.research.google.com/drive/1s0V0uTZrsHoEzmnGcvGM_MeX2NbPOI2U?usp=sharing#scrollTo=G_PdfMRZYGl6

But there is an error that I suspect is because of recent tensorflow/keras version changes in colab.

On the cell following the heading Random example, it fails on cnn.apply_model

FailedPreconditionError: 2 root error(s) found.
  (0) Failed precondition: Could not find variable conv2d/bias. This could mean that the variable has been deleted. In TF1, it can also mean the variable is uninitialized. Debug info: container=localhost, status=Not found: Container localhost does not exist. (Could not find resource: localhost/conv2d/bias)
	 [[{{node conv2d/BiasAdd/ReadVariableOp}}]]
	 [[batch_normalization_4/cond/Identity/_5]]
  (1) Failed precondition: Could not find variable conv2d/bias. This could mean that the variable has been deleted. In TF1, it can also mean the variable is uninitialized. Debug info: container=localhost, status=Not found: Container localhost does not exist. (Could not find resource: localhost/conv2d/bias)
	 [[{{node conv2d/BiasAdd/ReadVariableOp}}]]
0 successful operations.
0 derived errors ignored.

I was able to get past this error with another version of tensorflow/keras, adding the following to the top of the notebook:

!pip uninstall keras-nightly
!pip uninstall -y tensorflow
!pip install h5py==2.10.0  yy
!pip install tensorflow==2.4.0
!pip install keras==2.4.3

But the next error was on the cell following CAM for random example, failing on class_activation.run_gradcam.

AttributeError: module 'tensorflow' has no attribute 'get_default_session'

I have been trying a few other tensorflow/keras version combos without success.

If you happen to know a quick fix, I would greatly appreciate it.