The following are usage considerations:
- TD_SMOTE shows decreased throughput for higher concurrency for very large moderately imbalanced datasets. The function should be run at higher concurrencies with caution.
- TD_SMOTE's internal query processing requires a complex tree structure processing by Optimizer, which needs larger memory to handle larger number of columns. To increase that memory, dbscontrol flag MaxParseTreeSegs should be set to 12000.
- TD_SMOTE supports a maximum of 1000 columns for smote, and 500 for adasyn, borderline and smotenc (along with setting MaxParseTreeSegs to 12000). For smotenc, the total number of columns include InputColumns and CategoricalInputColumns.
This table provides information about errors and how to recover from them.
| Problem | Solution |
|---|---|
| Insufficient memory to parse this request. Optimizer did not have enough memory to parse the internal query. |
Reduce the number of columns in the input table or if TD_SMOTE is executed in nested call, UNION ALL, MINUS ALL operations, use TD_SMOTE as a standalone query. |
| Error in function TD_SMOTE: No synthetic samples were constructed because of no neighbors found in majority class. This error may happen for some combinations of input table and minority class where ADASYN algorithm may not be able to find any neighbor to sample from. |
Use SamplingStrategy of smote or increase the NumberOfNeighbors. |
| Error in function TD_SMOTE: No synthetic samples were constructed because of no examples found in border group for this class. This error may happen for some combinations of input table and minority class where BORDERLINE algorithm may not be able to find any sample in the border group. Samples in the border group are minority samples closer to the boundary with the majority class. |
Use SamplingStrategy of smote or increase the NumberOfNeighbors. |
| Error in function TD_OrdinalEncodingTransform: Category not found in FIT Table. Use DefaultValue argument while generating FIT Table for unknown categories in Input Table. This error may happen when EncodingsTable is generated using a subset of data from InputTable. |
Ensure that at least the rows from MinorityClass are in the InputTable and are used to create the EncondingsTable with Approach ('AUTO') and without DefaultValue argument. In short, use the same rows in both: EncodingsTable and InputTable. |