The numeric replacement is the exact observed-column mean.

highlighted = computed this step

Use only observed ages

The observed ages are 2, 6, and 10. The missing age is not included in the mean calculation.

observed ages=(2,6,10)\text{observed ages}=(2,6,10)
Missing Values ExactlyExact missing-cell replacements from a pinned policy.explicit missing-value policyrowage rawcolor rawage filledcolor filledage_missingcolor_missingA2red2red00B?blue6blue10C6?6red01D10red10red00age mean: (2+6+10)/3 = 6color mode: red count 2, blue count 1add missingness indicatorsexplicit preprocessing policy for displayed missing cellsnumeric uses observed mean; categorical uses observed modemissingness indicators stay visibleNOT training; NOT learning; NOT generalizationNOT accuracy; NOT calibration; NOT a guarantee of model qualityNOT probability truth

Compute the exact mean

The sum is 18 across 3 observed ages, so the mean is 6.

2+6+103=6{2 + 6 + 10\over3} = 6
Missing Values ExactlyExact missing-cell replacements from a pinned policy.explicit missing-value policyrowage rawcolor rawage filledcolor filledage_missingcolor_missingA2red2red00B?blue6blue10C6?6red01D10red10red00age mean: (2+6+10)/3 = 6color mode: red count 2, blue count 1add missingness indicatorsexplicit preprocessing policy for displayed missing cellsnumeric uses observed mean; categorical uses observed modemissingness indicators stay visibleNOT training; NOT learning; NOT generalizationNOT accuracy; NOT calibration; NOT a guarantee of model qualityNOT probability truth

Pin the numeric policy

The numeric policy is explicit: age missing uses the observed mean 6.

age ? uses 6\text{age } ? \text{ uses }6
Missing Values ExactlyExact missing-cell replacements from a pinned policy.explicit missing-value policyrowage rawcolor rawage filledcolor filledage_missingcolor_missingA2red2red00B?blue6blue10C6?6red01D10red10red00age mean: (2+6+10)/3 = 6color mode: red count 2, blue count 1add missingness indicatorsexplicit preprocessing policy for displayed missing cellsnumeric uses observed mean; categorical uses observed modemissingness indicators stay visibleNOT training; NOT learning; NOT generalizationNOT accuracy; NOT calibration; NOT a guarantee of model qualityNOT probability truth