Hi,
Thanks a lot for sharing your code on Bayesian NN, it sure was very useful. However, when experimenting on regression problems > 2 variables and classification problems, the model does not seem to learn anything. The loss pretty much remains constant after a few iterations. This is especially true for Relu activation. To show this, I am attaching a zip file where, when you run the file example_regression.py, you get the following plot.
[BayesianNN_Problem.zip] (https://github.com/krasserm/bayesian-machine-learning/files/3607763/BayesianNN_Problem.zip)
As you can see, the model does not learn the shape when we have 2 features. This actually changes, if I change the activation from ‘relu’ to ‘tanh’! any reason why?
Additionally, I also tried your model for classifying fashion mnist data, and unfortunately, the model does not learn anything here and simply produces an accuracy of 10%
I am not sure as to what is wrong here. Any help is greatly appreciated!
Thanks
Vineeth
Hi,
Thanks a lot for sharing your code on Bayesian NN, it sure was very useful. However, when experimenting on regression problems > 2 variables and classification problems, the model does not seem to learn anything. The loss pretty much remains constant after a few iterations. This is especially true for Relu activation. To show this, I am attaching a zip file where, when you run the file example_regression.py, you get the following plot.
[BayesianNN_Problem.zip] (https://github.com/krasserm/bayesian-machine-learning/files/3607763/BayesianNN_Problem.zip)
As you can see, the model does not learn the shape when we have 2 features. This actually changes, if I change the activation from ‘relu’ to ‘tanh’! any reason why?
Additionally, I also tried your model for classifying fashion mnist data, and unfortunately, the model does not learn anything here and simply produces an accuracy of 10%
I am not sure as to what is wrong here. Any help is greatly appreciated!
Thanks
Vineeth