Variable Specifications

[Samuwhale]

Had issues with rebasing, so I figured the easiest was to just create a new branch and add var_spec there. This PR replaces the old one #161.

It's still WIP, hence the draft.

Here's an example of how to use it:

from metasynth.distribution import FakerDistribution, DiscreteUniformDistribution, RegexDistribution
from metasynth.spec import MetaFrameSpec, VariableSpec

demo_file_path = demo_file()
demo_types={
    "Sex": pl.Categorical,
    "Embarked": pl.Categorical
}
df = pl.read_csv(demo_file_path, try_parse_dates=True, dtypes=demo_types)

# Old approach

df_spec_0 = {
    "PassengerId": {"unique": True},
    "Name": {"distribution": FakerDistribution("name")},
    "Fare": {"distribution": "ExponentialDistribution"},
    "Age": {
        "distribution": DiscreteUniformDistribution(20, 40),
    },
    "Cabin": {"distribution": RegexDistribution(r"[ABCDEF]\d{2,3}")}
}

mf_0 = MetaFrame.fit_dataframe(df, spec=df_spec_0)

print(mf_0.synthesize(10))

# New approach (option 1: creating VariableSpec objects directly, and passing a dictionary of those to fit_dataframe())
passenger_spec = VariableSpec()
passenger_spec.unique = True

name_spec = VariableSpec()
name_spec.distribution = FakerDistribution("name")

fare_spec = VariableSpec()
fare_spec.distribution = "ExponentialDistribution"

age_spec = VariableSpec()
age_spec.distribution = DiscreteUniformDistribution(600, 1000)

cabin_spec = VariableSpec()
cabin_spec.distribution = RegexDistribution(r"[ABCDEF]\d{2,3}")

df_spec_1 = {
    "PassengerId": passenger_spec.to_dict(),
    "Name": name_spec.to_dict(),
    "Fare": fare_spec.to_dict(),
    "Age": age_spec.to_dict(),
    "Cabin": cabin_spec.to_dict()
}

mf_1 = MetaFrame.fit_dataframe(df, spec=df_spec_1)

print(mf_1.synthesize(10))

# New approach (option 2: creating a MetaFrameSpec object, and passing that to fit_dataframe())

df_spec_2 = MetaFrameSpec(df)
df_spec_2["PassengerId"].unique = True
df_spec_2["Name"].distribution = FakerDistribution("name")
df_spec_2["Fare"].distribution = "ExponentialDistribution"
df_spec_2["Age"].distribution = DiscreteUniformDistribution(600, 1000)
df_spec_2["Cabin"].distribution = RegexDistribution(r"[ABCDEF]\d{2,3}")

mf_2 = MetaFrame.fit_dataframe(df, spec=df_spec_2.to_dict())

print(mf_2.synthesize(10))

[Samuwhale]

@qubixes and I discussed that it might be nice to set these specs through config files (probably TOML is nice)

[qubixes]

I think given the new PR #227, we can close this PR for now. Thanks for the inspiration!