Input Validator (diagnostics.input_validator)

Utilities for validating input data before modelling. These checks help ensure data quality and prevent common issues prior to model training.

Key checks

• check_nans(dataframe, target_col, media_cols, control_cols)

  • Verifies that specified columns contain no NaN values.

  • Raises ValueError with a list of offending columns.

• check_duplicate_columns(dataframe)

  • Ensures column names are unique.

  • Raises ValueError if duplicates are found.

• check_date_column(date_series, config)

  • Validates chronology, frequency, missing dates, and weekly start day.

  • Attempts to parse using config.get('date_format') if provided.

  • Raises ValueError if unsorted, irregular, or gaps are detected.

• check_column_variance(dataframe, columns, check_zeros_only=False)

  • Detects columns with zero variance (or all zeros when check_zeros_only=True).

  • Raises ValueError listing columns with issues.

Usage example

import pandas as pd
from src.diagnostics.input_validator import (
    check_nans,
    check_duplicate_columns,
    check_date_column,
    check_column_variance,
)

# Example inputs
config = {"date_format": None}
media_cols = ["tv_spend", "search_spend"]
control_cols = ["price", "competitor_index"]

# 1) Duplicate columns
check_duplicate_columns(df)

# 2) Date column integrity
check_date_column(df["date"], config)

# 3) NaNs across core columns
check_nans(df, target_col="revenue", media_cols=media_cols, control_cols=control_cols)

# 4) Zero-variance checks
check_column_variance(df, columns=media_cols + control_cols, check_zeros_only=False)

Notes

• These validators print structured status messages; errors raise ValueError and are intended to fail fast.

• For MMM-specific diagnostics (stationarity, VIF, transfer entropy), see the Pre‑Diagnostics Guide.