Multiple failures to install using Poetry and pip on Windows...

[richlysakowski]

System Info

Windows 10 Professional
Python 3.10, Python 3.12
PandasAI latest version 2.4.0

🐛 Describe the bug

Hey Guys, what is with the major incompatibilities of PandasAI with Pandas?

WHAT I TRIED:

I tried installing on 3 different Windows Professional machines:

1. in a conda environment, with pip install pandasai,
2. with poetry alone, with pip install pandasai,
3. with poetry tools in conda, then pip install.

THE RESULT I GOT WAS:

It failed on every installation, because of pip or poetry trying to compile the old 1.5.3 version of pandas.

You need to get caught up with Pandas. The package is called PANDAS AI.
This should be really obvious, but start with the latest Pandas version and compile AI APIs on top of it.

Dump Poetry and deliver conda packages. Don't make it hard for users and give us only pip-based packages to work with. Many users don't know how to install the Microsoft C++ 14 Build tools; this is why it failed to build in the first installation attempt using Poetry. Use conda to take away the burden of compilation expertise away from end users and puts it squarely onto developers.

Pandas 1.5.3 was released almost 2 years ago. Get with the program, PLEASE!! Please stay up to date with Pandas as a high priority.

[ZadeFrontier]

They keep mentioning that OpenAI requires pandas=1.5.3 and has a dependancy on it instead of trying to resolve the issue. I don't understand why they have to be inter-dependant on each other. They can be 2 different components and use agents to connect to each other.

[ZadeFrontier]

They also keep mentioning to wait till pandasai Version 3 to catch up to Pandas>2.

[richlysakowski]

They keep mentioning that OpenAI requires pandas=1.5.3 and has a dependancy on it instead of trying to resolve the issue. I don't understand why they have to be inter-dependant on each other. They can be 2 different components and use agents to connect to each other.

OpenAI, CrewAI, Llama index, Anthropic can use Pandas 2.2 without problems, so it is unclear why the PandasAI team has waited, other than resource constraints on their developer team.

[gdcsinaptik]

@richlysakowski we confirm that the upcoming release of PandasAI 3.0 fixes the issue with Pandas 2+ dependency. The release will also make the installation more lightweight. Overall, we have carefully worked for a major improvement of the library in terms of simplicity of use, not only installation. Thanks for your patience on this dependency issue.

P.S.
Looking forward to share a beta release of the 3.0 soon

[richlysakowski]

Will this be this month, next quarter, or next year (2026)? Surely you have some estimate of "v3.0" delivery. We need to know whether to expend any more time trying to use PandasAI "as-is". Right now PandasAI is very delicate and buggy with landmines in unexpected places. The setup is the most time-consuming part. Or move to another AI-powered dataframe mangler for Excel and DB tables until this open-source product is ready.

Being in honest communication with your user base is important.

Will there be a conda-based package installer for PandasAI?
Poetry is too problematic for end users in data science. venv is too risky for production.

---

***I found a partial answer to my own question... from @gdcsinaptik @gdcsinaptik. "It's coming soon (Q1 25), supporting pandas > 2, simpler to use, lighter and enhanced :)" ***

So, sometime in Q1-2025. At least there is a committed upper bound in time.