No Test Row in Fit

Fit Min and Max on Train

Compute preprocessing parameters from train rows only.

highlighted = computed this step

Fit the train-only minimum

Train rows A and B have x values 0 and 10. The train-only minimum is 0.

min⁡train=0\min_{\text{train}}=0
Train Only Preprocessing ExactlyExact min-max fit with train rows and full-data contrast.role tablerowrolexAtrain0Btrain10Ctest20fit parametersfitminmaxtrain-only A,B010full-data A,B,C contrast020train-only min-max fit and full-data contrastrowxtrain-only scaledfull-data scaledA000B1011/2C2021test row C changes the full-data fit; contrast onlyfit preprocessing on train rows only; full-data is contrasttest row changes numbers when it enters the fitdoes not claim: NOT leakage guarantee; NOT future performanceNOT accuracy; NOT generalization; NOT calibrated; NOT safeNOT best preprocessing; NOT probability truth

Fit the train-only maximum

The train-only maximum is 10.

max⁡train=10\max_{\text{train}}=10
Train Only Preprocessing ExactlyExact min-max fit with train rows and full-data contrast.role tablerowrolexAtrain0Btrain10Ctest20fit parametersfitminmaxtrain-only A,B010full-data A,B,C contrast020train-only min-max fit and full-data contrastrowxtrain-only scaledfull-data scaledA000B1011/2C2021test row C changes the full-data fit; contrast onlyfit preprocessing on train rows only; full-data is contrasttest row changes numbers when it enters the fitdoes not claim: NOT leakage guarantee; NOT future performanceNOT accuracy; NOT generalization; NOT calibrated; NOT safeNOT best preprocessing; NOT probability truth

Keep the fitted range

The fitted train-only range is 10 minus 0 = 10.

10−0=1010 - 0 = 10
Train Only Preprocessing ExactlyExact min-max fit with train rows and full-data contrast.role tablerowrolexAtrain0Btrain10Ctest20fit parametersfitminmaxtrain-only A,B010full-data A,B,C contrast020train-only min-max fit and full-data contrastrowxtrain-only scaledfull-data scaledA000B1011/2C2021test row C changes the full-data fit; contrast onlyfit preprocessing on train rows only; full-data is contrasttest row changes numbers when it enters the fitdoes not claim: NOT leakage guarantee; NOT future performanceNOT accuracy; NOT generalization; NOT calibrated; NOT safeNOT best preprocessing; NOT probability truth