Single decision trees support millions of attributes. Because the database cannot have millions of columns, you must spread the attributes across rows in the form of key-value pairs, where key is the name of the attribute and value is the value of the attribute. The Unpivoting function is useful for this purpose (see Unpivoting Example 1: Specified Target Columns, Default Optional Values).
| Table | Description |
|---|---|
| InputTable | [Required if you omit AttributeTableName and ResponseTableName, disallowed otherwise.] Contains attribute names and values and response values. |
| AttributeTableName | [Required if you omit InputTable, disallowed otherwise.] Contains attribute names and values. |
| ResponseTableName | [Required if you omit InputTable, disallowed otherwise.] Contains response values. |
| SplitsTable | [Optional] Contains user-specified splits. Every attribute in AttributeTableName must be in non-empty partition in this table. |
| CategoricalAttributeTableName | [Optional] Contains categorical attributes. |
The function ignores input rows with NULL values.
InputTable Schema
| Column | Data Type | Description |
|---|---|---|
| id_column | Any | Data point identifier. Cannot be NULL. |
| attribute_name_column | VARCHAR | Attribute name. Cannot be NULL. |
| attribute_value_column |
Numeric attribute: NUMERIC, INTEGER, BIGINT, or DOUBLE PRECISION Categorical attribute: Any |
Attribute value. If NULL, function estimates value by arithmetic means on an attribute basis. If estimate is out of range, function cannot use it to partition training data, so it is useless. |
| response_column | NUMERIC, INTEGER, BIGINT, or DOUBLE PRECISION | Response value for data point. Can be NULL. |
| weight_column | DOUBLE PRECISION | [Column appears only with Weighted ('true').] Weight of data point. Cannot be NULL. |
| actual_label | VARCHAR | Actual label of data point. |
AttributeTableName Schema
| Column | Data Type | Description |
|---|---|---|
| id_column | Any | Data point identifier. Cannot be NULL. |
| attribute_name_column | VARCHAR | Attribute name. Cannot be NULL. |
| attribute_value_column |
Numeric attribute: NUMERIC, INTEGER, BIGINT, or DOUBLE PRECISION Categorical attribute: Any |
Attribute value. If NULL, function estimates value by arithmetic means on an attribute basis. If estimate is out of range, function cannot use it to partition training data, so it is useless. |
| actual_label | VARCHAR | Actual label of data point. |
ResponseTableName Schema
| Column | Data Type | Description |
|---|---|---|
| id_column | Any | Data point identifier. Cannot be NULL. |
| response_column | NUMERIC, INTEGER, BIGINT, or DOUBLE PRECISION | Response value for data point. Can be NULL. |
| weight_column | DOUBLE PRECISION | [Column appears only with Weighted ('true').] Weight of data point. Cannot be NULL. |
The response table must not have a column named node_id.
SplitsTable Schema
| Column | Data Type | Description |
|---|---|---|
| attribute_name_column | VARCHAR | Attribute name. Cannot be NULL. Every attribute in AttributeTableName must be in non-empty partition in this table. |
| split_id | INTEGER | Split identifier. Cannot be NULL. |
| splits_valcol | NUMERIC, INTEGER, BIGINT, or DOUBLE PRECISION | Split value. Cannot be NULL. |
CategoricalAttributeTable Schema
| Column | Data Type | Description |
|---|---|---|
| attribute | VARCHAR | Categorical attribute name. |