Haifeng Li

Feel free to ask questions by creating tickets.

I added `DataFrame.fillna()` that applies on `FloatVector` and `DoubleVector`.

Sure, we will add support for multi class. BTW, `smile.deep.metric` package precision and recall support macro-, micro-, and weighted metrics of multi-class data.

It is added to master branch now. See [tests](https://github.com/haifengl/smile/blob/master/core/src/test/java/smile/validation/metric/PrecisionTest.java) for examples.

The new feature has to change the class APIs so that we couldn't include it in a bug release such as 3.1.1. But it will be available in next major...

It doesn't work on Windows 10. It works on Ubuntu 22.04 and macOS 13/14. It doesn't work on older macOS though (e.g. macOS 11). It is fine to drop the...

 `c10`, `asmjit`, and `libomp140` are missing on Windows 10.

On macOS 11, it cannot load `macosx-x86_64/libjnitorch.dylib`. However, the libjnitorch.dylib file does exist. ``` Caused by: java.lang.UnsatisfiedLinkError: Can't load library: /Users/xyz/.javacpp/cache/pytorch-2.2.2-1.5.11-20240502.145852-21-macosx-x86_64.jar/org/bytedeco/pytorch/macosx-x86_64/libjnitorch.dylib ```

It is okay to drop supporting macOS 11. Does PyTorch 2.2.2-1.5.11-SNAPSHOT support macOS on arm64 yet?

Thanks. It runs on Windows now. However, it doesn't produce good models on Windows though. I will create a new ticket about it.