The ML_TRAIN
routine provides
advanced options to influence model selection and training.
-
The
model_list
option permits specifying the type of model to be trained. If more than one type of model specified, the best model type is selected from the list. For a list of supported model types, see Section 3.16.13, “Model Types”. This option cannot be used together with theexclude_model_list
option.The following example trains either an
XGBClassifier
orLGBMClassifier
model.mysql> CALL sys.ML_TRAIN('heatwaveml_bench.census_train', 'revenue', JSON_OBJECT('task','classification', 'model_list', JSON_ARRAY('XGBClassifier', 'LGBMClassifier')), @census_model);
-
The
exclude_model_list
option specifies types of models that should not be trained. Specified model types are excluded from consideration. For a list of model types you can specify, see Section 3.16.13, “Model Types”. This option cannot be used together with themodel_list
option.The following example excludes the
LogisticRegression
andGaussianNB
models.mysql> CALL sys.ML_TRAIN('heatwaveml_bench.census_train', 'revenue', JSON_OBJECT('task','classification', 'exclude_model_list', JSON_ARRAY('LogisticRegression', 'GaussianNB')), @census_model);
-
The
optimization_metric
option specifies a scoring metric to optimize for. See: Section 3.16.14, “Optimization and Scoring Metrics”.The following example optimizes for the
neg_log_loss
metric.mysql> CALL sys.ML_TRAIN('heatwaveml_bench.census_train', 'revenue', JSON_OBJECT('task','classification', 'optimization_metric', 'neg_log_loss'), @census_model);
-
The
exclude_column_list
option specifies feature columns to exclude from consideration when training a model.The following example excludes the
'age'
column from consideration when training a model for thecensus
dataset.mysql> CALL sys.ML_TRAIN('heatwaveml_bench.census_train', 'revenue', JSON_OBJECT('task','classification', 'exclude_column_list', JSON_ARRAY('age')), @census_model);