Update tensorflow-probability version range

[williamjamir]

Loosening the requirements allows for the installation of the newer version 0.25 of tensorflow-probability, enabling the installation of tensorflow 2.18 and consequently numpy 2.0.

[williamjamir]

Hi @WillianFuks!

Currently setup.py specifies upper version bounds for Pandas and Tensorflow-probilitistic.

IMHO, strict upper bounds can lead to dependency conflicts and make it harder for users to integrate the library with other packages. To improve compatibility and the overall developer experience, it might be helpful to remove these upper version constraints.

This blog post from Henry Schreiner (core maintenair of many scientific building packages) explains the drawbacks of tight version bounds in more detail.

What do you think about loosening these constraints, I can open a PR removing them if you agree :)

[WillianFuks]

Hi @williamjamir

I certainly agree with you, I tried avoiding setting upper bounds for quite a while but then each new release from dependencies that contained backwards incompatible changes kept breaking things up.

It became a problem of finding a balance between avoiding conflicts internally and also externally with the user`s own packages versions installed.

So the strategy taken, in a way, is to keep up-to-date with current releases while protecting the code against new breaking changes. Not sure if that was the best approach, it was what worked at the time.

I'll read your reference article, if there are any ideas on how to remove the upper bounds while avoiding having the code breaking unexpectedly I'm certainly in favor of doing that.