Henry Leung

> hello, today google release tensorflow 2.0, but my pc have not install it because i using cuda 10.1, please build a 10.1 version for users.thank you btw there is...

> I build tf2.0 with GPU support successfully.I use it with keras,strangely,it slower then tf1.14. > Some code don't work in tf2.0,I couldn't use it. I have tried but I...

> Without CuDNNLSTM, performance dipped 3-5 times. CPU load increased 2 times (70% vs 35%). > How to replace CuDNNLSTM correctly? > LSTM is not a good replacement option. >...

> 官方放棄cmake了 我想我發PR大概也不會被接受 Does it mean you will not build Tensorflow 1.10 anymore because they have dropped cmake support?? I guess getting Bazel to work on Windows is still painful...

Yes ``ApogeeBCNN`` use 1D convolution, you should use ``MNIST_BCNN`` for images. i.e. ```python from astroNN.models import MNIST_BCNN ```

For example see here: https://github.com/henrysky/astroNN/blob/master/demo_tutorial/NN_uncertainty_analysis/Uncertainty_Demo_MNIST.ipynb

I think I actually have fixed the issue (as you can see the notebook was updated earlier this year and no issue was observed, uncertainty looks reasonable). You are using...

do you have a notebook or code to show what have you been doing? it is very weird that ``output_categorical_accuracy``, ``val_output_categorical_accuracy`` are always 1.

One immediate issue I have noticed is your images are ``astype(int)`` where they should be float

Sorry for the late reply. I have add a function as a first step to solve your issue. So now this new function `transfer_weights()` should transfer all the weights to...