Explicit Policy

Missing Cells Stay Visible

Start with the raw table: missing cells are shown, not hidden.

highlighted = computed this step

Start with visible missing cells

The displayed table has one numeric feature, age, and one categorical feature, color. Missing cells stay visible as question marks.

raw cells include ?\text{raw cells include } ?
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

Row B is missing age

Row B has age missing and color blue. The missing age is not guessed yet; the policy comes next.

Bage=?B_{\text{age}} = ?
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

Row C is missing color

Row C has observed age 6 and color missing. The missing marker stays visible in the raw side of the table.

Cage=6,Ccolor=?C_{\text{age}}=6,\quad C_{\text{color}}=?
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