Dirk T.

@jsimons8 You could build a Functor that takes another Functor `f` and computes the gradients for it using [Finite Differences](https://en.wikipedia.org/wiki/Finite_difference). ```C++ template class NumDiff { private: const Functor f; const...

Two possibilities: Either throwing Exceptions: ``` java Var var = simpleVar(1f); Val val = var.map( x -> x*x ); val.addChangeProposedListener( (oldVal,newVal) -> { if(newVal > 3) throw new IllegalArgumentException(); }...

In 95% of all cases that is absolutely true, but: - Let's say it's a Bug. It may still be important that the Model stays intact. What if you need...

> You should validate all user input in the first place. The problem is that the validity of a text value depends on where it is in the whole data...

That would have been my idea for a Workaround as well. One concern I had with that is edge cases: Wouldn't values like `0` and `Integer.MAX_VALUE % list.size()` be tested...

FIY you can also match code point ranges. I had made a corresponding pull request a while ago (PR #35). The relevant function is [makeCodePointRange( int min, int max )](https://github.com/cs-au-dk/dk.brics.automaton/blob/c1b119fec95345b45839c0fce9e383bdd4cf0e63/src/main/dk/brics/automaton/BasicAutomata.java#L122-L213),...

For me that would be the Browser mostly. Outside the Browser there are many alternatives (TF4Py, PyTorch, ...) while inside the Browser I see none. My focus is on having...

It's good to hear that You're committed to extending `tf.linalg` and I realize how much review work is involved. The past few weeks, I've been trying to think of ways...

One minor downside of Python→JS ports is that they usually come with ginormous package sizes. Pyodide is ~12mb if I see it correctly (~15 times the size of `tf.min.js`). For...

Added it to the list :) If the matrices are small, You can already compute the determinant using the current implementation of the QR decomposition: ```js function det( /*tf.Tensor*/a )...