Request: relax dependency requirements

[Giqles]

The dependencies specified in setup.py are strict, some of which are pointing to relatively old versions of key analytics libraries. For example Pandas and Numpy are now on 2.x versions, but setup.py dependencies are locked to 1.x versions.

This makes it hard to adopt synthdid, as it places quite strong constraints on which other libraries we can use in the same environment. Relaxing these would make it much easier to build workflows around synthdid.

[y1my1]

yes, the same problem here, can we relax the dependency to not use fixed versions?

[damiancclarke]

Dependency version pins in setup.py cause conflicts with common packages

The current setup.py uses exact version pins (e.g. numpy==1.23.5, pandas==1.5.3, matplotlib==3.7.1, attrs==19.1.0) which makes it very difficult to use synthdid alongside other common econometrics packages in the same environment.

For example, packages like plotnine, rdrobust, numba, and jsonschema all require newer versions of these same libraries. When synthdid is installed, pip either downgrades those packages (breaking other tools) or raises dependency conflicts.

Suggested fix: Replace exact pins with minimum version constraints where possible. For example:

# Instead of:
"numpy==1.23.5",
"pandas==1.5.3",
"matplotlib==3.7.1",
"attrs==19.1.0",

# Consider:
"numpy>=1.23.5",
"pandas>=1.5.3",
"matplotlib>=3.7.1",
"attrs>=22.2.0",

I tested synthdid with newer versions of all these packages and the core functionality (estimation, vcov, plotting) works correctly - the exact pins don't appear to reflect genuine compatibility constraints.

Would the maintainers be open to a PR relaxing these pins? Alternatively, can I issue a PR with this?