Perform advanced model tuning in AutoML UI

warning

The advanced model tuning is available in the AutoML interface only from 10.2.9.2.

As you train a model in the AutoML interface, you can fine-tune it with additional parameters using the Advanced parameters JSON editor.

By default, the Advanced parameters editor shows the following parameters:

• executionProfile defines if evaluation results and metadata for them should be preserved (DEBUG mode). Possible values are PRODUCTION (default) and DEBUG.

• enableFeatureImportanceExplanation prints top N features, each with detailed info and score. Possible values are TRUE (default) and FALSE.

• extendedCrossValidation is a Boolean flag defining whether to apply extended cross-validation or not. Possible values are TRUE (default) and FALSE.

Below you can find more advanced model training and execution parameters you can set in the Advanced parameters editor depending on the model type.

Model training flags

Shared

Parameter	Parameter alias	Admissible values	Description
ProcessGlobalConst.TRAINING_SET_PREPARED	"trainingSetPrepared"	"TRUE" or "FALSE"; the default value is "FALSE".	Defines if a training set is already inside <egid>/input/training-set.
ConfigurationConstants.PARAM_COMPRESS_FEATURES	"compressFeatures"	"TRUE" or "FALSE"; the default value is "TRUE".	Defines if features are to be compressed into an archive or not.
ConfigurationConstants.GLOBAL_PARAM_FULL_EVAL_NUM_FOLDS	"fullEvalNumFolds"	For information extraction (IE), the default value is "5". For classification, it is "3".	Defines the number of folds for splitting a training set.
ConfigurationConstants.ML_CONFIG_TYPE	"mlConfig.type"	["LibLinear" , "VW"]; default: "LibLinear".	Defines the used ML algorithm.
ProcessGlobalConst.PARAM_EXECUTION_PROFILE	"executionProfile"	["PRODUCTION", "DEBUG"]; default: "PRODUCTION".	Defines if the work directory for the training should be preserved (DEBUG mode).
GenericPipelineConfiguration.MIN_FEATURES_FREQUENCY	"minFeaturesFrequency"	Default: 10.	Defines the minimum frequency of the feature required to include it in a model.
ConfigurationConstants.PARAM_MAX_EXPERIMENTS_IN_PARALLEL	"maxExperimentsInParallel"	Default: 8 .	Defines the maximum count of parallel experiments for selecting the algorithm parameter.
GenericPipelineConfiguration.NORMALIZE_SCORE	"normalizeScore"	"TRUE" or "FALSE"; defaults: for Liblinear, "TRUE"; for PythonClassifier, "FALSE"; for VowpalWabbitClassifier, "FALSE".	Defines the score normalization for the MlConfig classifier.
ConfigurationConstants.DATA_CLEANSING_MODE	"dataCleansing"	"TRUE" or "FALSE"; the default value is "FALSE".	Switches the composer to the mode when the model output comprises features used as the input data for the data cleansing tool.
ConfigurationConstants.ENABLE_FEATURE_IMPORTANCE_EXPLANATION	"enableFeatureImportanceExplanation"	"TRUE" or "FALSE"; the default value is "FALSE".	Prints the top N features, each with detailed info and score.

Information extraction

Parameter	Parameter alias	Admissible values	Description
ConfigurationConstants.PARAM_MAX_FIELDS_IN_PARALLEL	"fieldsInParallel"	Unlimited	Defines the maximum number of parallel submodels running in a single eval process.
ConfigurationConstants.GLOBAL_PARAM_PRE_EVAL_NUM_FOLDS	"preEvalNumFolds"	Default: for IE, "5"	Defines the number of folds in experiments.
ConfigurationConstants.GLOBAL_PARAM_USE_FULL_TRAINING_SET_FOR_PREEVAL	"useFullTrainSetForPreeval"	"TRUE" or "FALSE"; the default value is "FALSE".
ProcessGlobalConst.PARAM_USE_TOKENIZERS	"useTokenizers"	"TRUE" or "FALSE"; the default value is "TRUE".	Generates tokenizers based on the analysis of gold data in a training set.
ConfigurationConstants.PARAM_SKIP_DOCUMENTS_WITHOUT_TAG	"skipDocumentsWithoutTag"	"TRUE" or "FALSE"; the default value is "TRUE".	Defines whether to skip documents without tags or not.
ConfigurationConstants.PARAM_IGNORE_DOCUMENT_WITH_SAME_GOLDS	"ignoreSameDocument"	"TRUE" or "FALSE"; the default value is "TRUE".	Defines whether to ignore a document if it contains the same golds as in previous documents.
ConfigurationConstants.SEQUENCE_OF_OUTCOME_PROCESSOR_CLASS_NAME	"sequenceOfOutcomeProcessorClassName"	"com.workfusion.nlp.uima.annotator.ie.sequence.SequenceOfBIESOTagsLogprobProcessor", "com.workfusion.nlp.uima.annotator.ie.sequence.SequenceOfBIESOTagsProcessor", "com.workfusion.nlp.uima.annotator.ie.sequence.DefaultSequenceOfTagProcessor"; default: "com.workfusion.nlp.uima.annotator.ie.sequence.SequenceOfBIESOTagsProcessor".	Defines the full class name for the IE sequence processor.
ConfigurationConstants.PARAM_APPLY_CHECKER_MODEL_ON_EXECUTION	"applyCheckerModelOnExecution"	"TRUE" or "FALSE"; the default value is "FALSE".	Sets the usage of the checker model for execution.
ConfigurationConstants.PARAM_EXTENDED_CROSS_VALIDATION	"extendedCrossValidation"	"TRUE" or "FALSE"; the default value is "FALSE".	Defines whether to perform extended cross-validation or not.
GenericPipelineConfiguration.VALUE_BASED_EVALUATON	"valueBasedEvaluation"	"TRUE" or "FALSE"; the default value is "FALSE".	Defines the type of the processor to write the model results on the full eval stage.
GenericPipelineConfiguration.CASE_SENSITIVE	"caseSensitive"	"TRUE" or "FALSE"; the default value is "FALSE".	Defines the letter case normalization for ValueBasedWriteResultsProcessor to write the model results on the full-eval stage.
GenericPipelineConfiguration.PUNCT_INCLUDED	"puncCharPrefixSuffixIncluded"	"TRUE" or "FALSE"; the default value is "TRUE".	Defines the punctuation normalization for ValueBasedWriteResultsProcessor to write the model results on the full-eval stage.

Classification

Parameter	Parameter alias	Admissible values	Description
GenericPipelineConfiguration.BALANCE_WEIGHTS	"balanceWeights"	"TRUE", "FALSE"; the default value is "FALSE".	Enables the balanced mode. Uses the values of y to automatically adjust weights inversely proportional to class frequencies in the input data.
ConfigurationConstants.BALANCED_TRAINING_ARGUMENTS	"balancedTrainingArguments"	""	This is the parameter that defines the training arguments for a balanced training set.
ConfigurationConstants.PARAM_REBALANCE_TRAINING_SET	"rebalanceTrainingSet"	"TRUE", "FALSE"; the default value is "FALSE".	Defines whether to use a balanced training set or not.
ConfigurationConstants.PREFERRED_OUTCOMES_ORDER	"preferredOutcomesOrder"	"[]"	This is the parameter that defines the order of outcome classes.

Model execution flags

Shared

Parameter	Parameter alias	Admissible values	Description
ProcessGlobalConst.PARAM_EXECUTION_PROFILE	"executionProfile"	["PRODUCTION", "DEBUG"]	Defines if features are selected on extract and if evaluation results and metadata for them should be preserved (DEBUG mode).
GenericPipelineConfiguration.NORMALIZE_SCORE	"normalizeScore"	"TRUE","FALSE"; defaults: for Liblinear, "TRUE"; for PythonClassifier, "FALSE"; for VowpalWabbitClassifier, "FALSE".	Defines the score normalization for the MlConfig classifier.
ConfigurationConstants.ENABLE_FEATURE_IMPORTANCE_EXPLANATION	enableFeatureImportanceExplanation"	"TRUE","FALSE"; default: "FALSE"	Prints the top N features, each with detailed info and score.