I see that Kam Nasim mentioned my code rbf_keras. I have rewritten it yesterday to work with tensorflow 2.0 and added more examples of usage.

The new repo can be found at https://github.com/PetraVidnerova/rbf_for_tf2

Centres can be set up by k-means, not only randomly. Or you can define your own initializers for centres and widths.

Dear Dan Hurwitz,

Look the links, may be useful.

http://benjaminbolte.com/blog/2016/keras-language-modeling.html

http://machinelearningmastery.com/tutorial-first-neural-network-python-keras/

Regards, Shafagat

newrbf function in matlab

I too am curious to know the answer to this. Is there a native implementation of RBFs available in Keras similar to what they have for say CNNs?

actually came across this:

https://github.com/PetraVidnerova/rbf_keras

They are selecting the Centroids randomly, which is non-ideal... but this can be a good starting point and extended to incorporate some type of clustering (K-Means) for Centroid selection.

I'm curious about this topic since I haven't been able to come up ith some kind of a tutorial of some sort to implement this kind of method. Has anyone came up with something like that?

I think this article in medium.com will be very helpful and it provides an implementation in Keras.

https://medium.com/datadriveninvestor/building-radial-basis-function-network-with-keras-estimating-survivors-of-titanic-a06c2359c5d9

I see that Kam Nasim mentioned my code rbf_keras. I have rewritten it yesterday to work with tensorflow 2.0 and added more examples of usage.

The new repo can be found at https://github.com/PetraVidnerova/rbf_for_tf2

Centres can be set up by k-means, not only randomly. Or you can define your own initializers for centres and widths.