Custom training loop with tf probability Keras API

[Ghaith Habboub]

Hi,

I am hoping to expand a probabilistic Convolutional network to handle survival analysis using variation of the methods described here https://adamhaber.github.io/post/survival-analysis/  and by relying on Keras model as opposed to sequential joint distribution method.

The training will need to be custom training though to handle censored and non-censored data differently. Given inference and optimization is different here as opposed to the one described by non-probabilistic tensorflow Keras model here  https://www.tensorflow.org/tutorials/customization/custom_training_walkthrough

Any hint about the easiest way to do custom training and avoid the model.fit method?

Thanks,

Ghaith

Sent from my iPhone

[Krzysztof Rusek]

Recent Keras has train_step and predict_step methods for fine tuning training without a custom loop.

Maybe this will be the best for you as its as for me.

Take a look at my micro SVI library for Keras:

https://github.com/krzysztofrusek/keras_svi/blob/main/keras_svi/svi.py

Wiadomość napisana przez Ghaith Habboub <ghaith...@gmail.com> w dniu 08.12.2020, o godz. 14:31:

--
You received this message because you are subscribed to the Google Groups "TensorFlow Probability" group.
To unsubscribe from this group and stop receiving emails from it, send an email to tfprobabilit...@tensorflow.org.
To view this discussion on the web visit https://groups.google.com/a/tensorflow.org/d/msgid/tfprobability/3F0D9E72-0C32-41B4-ADE4-ACABA623DEA9%40gmail.com.

[Ghaith Habboub]

Thank you

This is definitely a familiar path I can work with.

Thanks,

Ghaith

--

Ghaith Habboub, MD

Neurosurgery Resident

Cleveland Clinic

Cleveland, Ohio