Description
Compare distinct entity support across two split labels.
Model validation depends on when information became available and how records or entities enter splits. A timestamp inequality or group overlap count can identify one local issue without proving absence of every form of leakage.
Compare distinct entity support across two split labels.
When to use this tool
Use Split Entity Support Gap when checking an explicit feature-time or split-membership condition.
How to use the tool and how it works
- Enter entityIds, splits, selectedSplit, referenceSplit. Keep parallel arrays aligned. Use the units and conventions below.
- Run the tool. Correct any format, range, or cross-field input error before interpreting the output.
- Read selectedEntities, referenceEntities, sharedEntities, selectedOnly. Use the output convention below.
Compare distinct entity support across two split labels. Require aligned arrays and different selected/reference labels; form sets and return their sizes, intersection count and selected-only count.1
| Input field | Meaning and units | Accepted input |
|---|---|---|
| entityIds | Entity Ids | Required. Array of text; maximum UTF-16 length 8192; maximum items 256. |
| splits | Splits | Required. Array of text; maximum UTF-16 length 8192; maximum items 256. |
| selectedSplit | Selected Split | Required. Text; maximum UTF-16 length 8192. |
| referenceSplit | Reference Split | Required. Text; maximum UTF-16 length 8192. |
| Output field | Interpretation | Returned shape |
|---|---|---|
| selectedEntities | Selected Entities | Required. Safe integer. |
| referenceEntities | Reference Entities | Required. Safe integer. |
| sharedEntities | Shared Entities | Required. Safe integer. |
| selectedOnly | Selected Only | Required. Safe integer. |
{
"input": {
"entityIds": [
"a",
"b",
"a"
],
"splits": [
"train",
"train",
"test"
],
"selectedSplit": "train",
"referenceSplit": "test"
},
"output": {
"selectedEntities": 2,
"referenceEntities": 1,
"sharedEntities": 1,
"selectedOnly": 1
}
} Limitations and assumptions
- Repeated records are deduplicated by entity ID; literal overlap alone is not a complete leakage assessment.
- Inputs must satisfy both the field types and the cross-field conditions above. Invalid input causes an error; a successful false flag or documented placeholder is a different result.
Alternative or Complementary methods
Use a complete data-provenance and validation-design review, including label horizons, entity dependence and preprocessing fit scope.
References
Similar or alternative tools
- Split Group Empty Fold Budget
Identify empty and below-minimum group-count folds.
- Data Split Chronological Cutoff Membership
Assign a timestamp to complementary chronological splits.