Usage

Regex Column Selection

If you have a DataFrame with many similar fields with similar names, you can select by a regex.

import pandas as pd
from recx import Rec, AbsTolCheck

df = pd.DataFrame({
    "metric_1": [0.1, 0.2, 0.3],
    "metric_2": [0.1, 0.2, 0.3],
    "metric_3": [0.1, 0.2, 0.3],
})

df2 = df.copy()
df2.loc[2, "metric_3"] = 0.6

rec = Rec({
    r"^metric_": AbsTolCheck(tol=0.01, regex=True),
})

rec.run(df, df2).summary()

Build in checks

Equal

from recx import EqualCheck

EqualCheck()

Use for a simple equality check.

Absolute Tolerance

from recx import AbsTolCheck

AbsTolCheck(tol=0.05)

This will fail when |baseline - candidate| > tol.

Relative Tolerance

from recx import RelTolCheck

RelTolCheck(tol=0.05)

This will fail when |baseline - candidate| / |candidate| > tol.

Custom Check

from recx.checks import ColumnCheck
import pandas as pd
import numpy as np

class SignCheck(ColumnCheck):
    def check(self, baseline: pd.Series, candidate: pd.Series):
        mask = np.sign(baseline) != np.sign(candidate)

        # Return a DataFrame of the bad rows. You can any
        # diagnostic columns you like.
        return pd.DataFrame({
            "baseline": baseline[mask],
            "candidate": candidate[mask],
        })

rec = Rec({
    "col": SignCheck(),
})

Date Alignment

A common task is comparing a time based dataset once it has been updated. In this situation, the new version (the candidate) many have extra rows with dates (or timestamps) ahead of the old version (the baseline).

You can use align_date_col="date_col" on the Rec class to skip these new dates.

rec = Rec(
    columns={...},
    align_date_col="date"  # Name of the date column
)

Skipping Columns

Assign None to a column key:

rec = Rec(
    columns={
        "not_important": None,
    },
)

or set check_all=False to ignore all unspecified columns:

rec = Rec(
    check_all=False,
)