An interesting experiment: https://medium.com/@zaleslaw/how-to-train-the-linear-regression-model-with-tensorflow-java-api-on-kotlin-89e2bf2f2b34

Nice. I think that kotlin builders on top of those operations could significantly simplify model building. We can work on that if someone would volonteer on explaingin what actions are actually needed. It seems it is even possible to use kmath expressions to build models. But I am not sure yet

tensorflow bundles keras which is a more idiomatic and cross-tool library. doing exlicit TF is like hand optimization in other code domains, you would only want that level of tooling if you knew of a feature that had no keras wrapper.

@jimn AFAIK, Keras-like Java Tensorflow API is still very much work in progress.

@jimn it will be, but later. We should light the graphs before presenting the Keras API that wraps this graph