Unpin requirements

[schlegelp]

Describe the use case that is addressed by this feature. Hi! This is not so much of a feature but rather to prevent incompatibilities/issues with other libraries: Currently, allensdk has rather narrow requirements: pandas, for example, is pinned at 0.25.1-0.25.3. The latest of those versions is almost 2 years old! The same goes for e.g. numpy where your latest allowed version is from June 2020. I have only spot checked other requirements and there are more that are dated.

Describe the solution you'd like Ideally, you would loosen the restrictions for most/all requirements such that you only define a minimum version. I would be very surprised if e.g. either the most recent numpy or pandas versions break any of your code.

Describe alternatives you've considered I appreciate that for production/dev/test environments strict requirements are useful. If that's your main concern, I would suggest writing a permissive requirements.txt for use in setup.py and for pip installs and provide an additional restrictive requirements-dev.txt for developers.

[tfardet]

Given that this is still an issue, I was wondering whether it has been discussed and specific problems where raised. Currently this really makes it difficult to let everyday users use the allensdk: the library install breaks for "no reasons" while it's actually still compatible with newer numpy and pandas versions. Right now I need to run it in a virtualenv (and often to additionally force pip to ignore version dependencies to make it work with other libraries that do need newer numpy versions).