For example, the documentation of TensorLike is located in tfa.image.color_ops:
https://www.tensorflow.org/addons/api_docs/python/tfa/image/color_ops/TensorLike
Presumably, the system selects the module where the symbol was first found. This is not a big issue but it produces strange and unexpected documentation for type annotations, see for example:
https://www.tensorflow.org/addons/api_docs/python/tfa/seq2seq/LuongAttention
I see 2 possible solutions:
- Hide non public symbols in the generated documentation
- Properly export type aliases in a public module