ML Prep

Supervised dataset

from metdatapy.mlprep import make_supervised
sup = make_supervised(df, targets=["temp_c"], horizons=[1, 3], lags=[1,2,3])

Adds {col}_lag{n} for numeric columns and targets like temp_c_t+1.

Time-safe splits

from metdatapy.mlprep import time_split
splits = time_split(sup, train_end=pd.Timestamp("2025-01-15T00:00Z"))

For proportion-based chronological splits, use time_split_by_fraction:

from metdatapy.mlprep import time_split_by_fraction

splits = time_split_by_fraction(
    sup,
    train=0.70,
    validation=0.15,
    test=0.15,
)

metadata = splits["metadata"]  # fractions and row counts

Both helpers preserve chronological order and return non-overlapping train, val, and test frames.

Scaling

from metdatapy.mlprep import fit_scaler, apply_scaler
scaler = fit_scaler(splits["train"], method="standard")
train_scaled = apply_scaler(splits["train"], scaler)
val_scaled = apply_scaler(splits["val"], scaler)
test_scaled = apply_scaler(splits["test"], scaler)