[BUG] `AttributeError` in `PandasDataFrame.__init__` with `triad>=0.9.2`

[charlesbluca]

[BUG] `AttributeError` in `PandasDataFrame.__init__` with `triad>=0.9.2`

[goodwanghan]

Ah, sorry, but I think Fugue side has finished the change to depend on dask-sql when the backend is dask, I will make a pr to dask-sql to remove the unnecessary part.

[paolorechia]

Hey, I'm also affected by this bug, using Spark backend.

Some relevant requirements:

adtk==0.6.2
pyod==1.0.1
numpy==1.22.3
fugue==0.8.3

My application suddenly stopped working from one deployment to another. It seems like the setup.py of fugue pins down versions in a rather risky way:

        "triad>=0.9.3",
        "adagio>=0.2.4",
        # sql dependencies
        "qpd>=0.4.4",
        "fugue-sql-antlr>=0.1.6",

So I'm guessing triad was updated, which caused my pinned down version of fugue to break. I see two long term solutions to this problem:

1. Change the pinning strategy of the fugue project, to something like this:

        "triad<=0.9.3",
        "adagio<=0.2.4",
        # sql dependencies
        "qpd<=0.4.4",
        "fugue-sql-antlr<=0.1.6",

This would ensure sub-dependencies are only updated with a specific version update on fugue.

2. Add a disclaimer in the installation steps that proper requirement locking is required from end-users, by using something like pipenv. Currently documentation recommends installation by pip which is a ticking bomb: https://github.com/fugue-project/fugue#installation