gle properties | IBM Cloud Pak for Data as a Service

gle properties

Last updated: Feb 11, 2025

gle properties

GLE node icon A GLE extends the linear model so that the target can have a non-normal distribution, is linearly related to the factors and covariates via a specified link function, and so that the observations can be correlated. Generalized linear mixed models cover a wide variety of models, from simple linear regression to complex multilevel models for non-normal longitudinal data.

Table 1. gle properties
`gle` Properties	Values	Property description
`custom_target`	flag	Indicates whether to use target defined in upstream node (`false`) or custom target specified by `target_field` (`true`).
`target_field`	field	Field to use as target if `custom_target` is `true`.
`use_trials`	flag	Indicates whether additional field or value specifying number of trials is to be used when target response is a number of events occurring in a set of trials. Default is `false`.
`use_trials_field_or_value`	`Field` `Value`	Indicates whether field (default) or value is used to specify number of trials.
`trials_field`	field	Field to use to specify number of trials.
`trials_value`	integer	Value to use to specify number of trials. If specified, minimum value is 1.
`use_custom_target_reference`	flag	Indicates whether custom reference category is to be used for a categorical target. Default is `false`.
`target_reference_value`	string	Reference category to use if `use_custom_target_reference` is `true`.
`dist_link_combination`	`NormalIdentity` `GammaLog` `PoissonLog` `NegbinLog` `TweedieIdentity` `NominalLogit` `BinomialLogit` `BinomialProbit` `BinomialLogC` `CUSTOM`	Common models for distribution of values for target. Choose `CUSTOM` to specify a distribution from the list provided by `target_distribution`.
`target_distribution`	`Normal` `Binomial` `Multinomial` `Gamma` `INVERSE_GAUSS` `NEG_BINOMIAL` `Poisson` `TWEEDIE` `UNKNOWN`	Distribution of values for target when `dist_link_combination` is `Custom`.
`link_function_type`	`UNKNOWN` `IDENTITY` `LOG` `LOGIT` `PROBIT` `COMPL_LOG_LOG` `POWER` `LOG_COMPL` `NEG_LOG_LOG` `ODDS_POWER` `NEG_BINOMIAL` `GEN_LOGIT` `CUMUL_LOGIT` `CUMUL_PROBIT` `CUMUL_COMPL_LOG_LOG` `CUMUL_NEG_LOG_LOG` `CUMUL_CAUCHIT`	Link function to relate target values to predictors. If `target_distribution` is `Binomial` you can use: `UNKNOWNIDENTITYLOGLOGITPROBITCOMPL_LOG_LOGPOWERLOG_COMPLNEG_LOG_LOGODDS_POWER` If `target_distribution` is `NEG_BINOMIAL` you can use: `NEG_BINOMIAL` If `target_distribution` is `UNKNOWN`, you can use: `GEN_LOGITCUMUL_LOGITCUMUL_PROBITCUMUL_COMPL_LOG_LOGCUMUL_NEG_LOG_LOGCUMUL_CAUCHIT`
`link_function_param`	number	Tweedie parameter value to use. Only applicable if `normal_link_function` or `link_function_type` is `POWER`.
`tweedie_param`	number	Link function parameter value to use. Only applicable if `dist_link_combination` is set to `TweedieIdentity`, or `link_function_type` is `TWEEDIE`.
`use_predefined_inputs`	flag	Indicates whether model effect fields are to be those defined upstream as input fields (`true`) or those from `fixed_effects_list` (`false`).
`model_effects_list`	structured	If `use_predefined_inputs` is `false`, specifies the input fields to use as model effect fields.
`use_intercept`	flag	If `true` (default), includes the intercept in the model.
`regression_weight_field`	field	Field to use as analysis weight field.
`use_offset`	`None` `Value` `Variable`	Indicates how offset is specified. Value `None` means no offset is used.
`offset_value`	number	Value to use for offset if `use_offset` is set to `offset_value`.
`offset_field`	field	Field to use for offset value if `use_offset` is set to `offset_field`.
`target_category_order`	`Ascending` `Descending`	Sorting order for categorical targets. Default is `Ascending`.
`inputs_category_order`	`Ascending` `Descending`	Sorting order for categorical predictors. Default is `Ascending`.
`max_iterations`	integer	Maximum number of iterations the algorithm will perform. A non-negative integer; default is 100.
`confidence_level`	number	Confidence level used to compute interval estimates of the model coefficients. A non-negative integer; maximum is 100, default is 95.
`test_fixed_effects_coeffecients`	`Model` `Robust`	Method for computing the parameter estimates covariance matrix.
`detect_outliers`	flag	When true the algorithm finds influential outliers for all distributions except multinomial distribution.
`conduct_trend_analysis`	flag	When true the algorithm conducts trend analysis for the scatter plot.
`estimation_method`	`FISHER_SCORING` `NEWTON_RAPHSON` `HYBRID`	Specify the maximum likelihood estimation algorithm.
`max_fisher_iterations`	integer	If using the `FISHER_SCORING` `estimation_method`, the maximum number of iterations. Minimum 0, maximum 20.
`scale_parameter_method`	`MLE` `FIXED` `DEVIANCE` `PEARSON_CHISQUARE`	Specify the method to be used for the estimation of the scale parameter.
`scale_value`	number	Only available if `scale_parameter_method` is set to `Fixed`.
`negative_binomial_method`	`MLE` `FIXED`	Specify the method to be for the estimation of the negative binomial ancillary parameter.
`negative_binomial_value`	number	Only available if `negative_binomial_method` is set to `Fixed`.
`use_p_converge`	flag	Option for parameter convergence.
`p_converge`	number	Blank, or any positive value.
`p_converge_type`	flag	True = Absolute, False = Relative
`use_l_converge`	flag	Option for log-likelihood convergence.
`l_converge`	number	Blank, or any positive value.
`l_converge_type`	flag	True = Absolute, False = Relative
`use_h_converge`	flag	Option for Hessian convergence.
`h_converge`	number	Blank, or any positive value.
`h_converge_type`	flag	True = Absolute, False = Relative
`max_iterations`	integer	Maximum number of iterations the algorithm will perform. A non-negative integer; default is 100.
`sing_tolerance`	integer
`use_model_selection`	flag	Enables the parameter threshold and model selection method controls..
`method`	`LASSO` `ELASTIC_NET` `FORWARD_STEPWISE` `RIDGE`	Determines the model selection method, or if using `Ridge` the regularization method, used.
`detect_two_way_interactions`	flag	When `True` the model will automatically detect two-way interactions between input fields. This control should only be enabled if the model is main effects only (that is, where the user has not created any higher order effects) and if the `method` selected is Forward Stepwise, Lasso, or Elastic Net.
`automatic_penalty_params`	flag	Only available if model selection `method` is Lasso or Elastic Net. Use this function to enter penalty parameters associated with either the Lasso or Elastic Net variable selection methods. If `True`, default values are used. If `False`, the penalty parameters are enabled custom values can be entered.
`lasso_penalty_param`	number	Only available if model selection `method` is Lasso or Elastic Net and `automatic_penalty_params` is `False`. Specify the penalty parameter value for Lasso.
`elastic_net_penalty_param1`	number	Only available if model selection `method` is Lasso or Elastic Net and `automatic_penalty_params` is `False`. Specify the penalty parameter value for Elastic Net parameter 1.
`elastic_net_penalty_param2`	number	Only available if model selection `method` is Lasso or Elastic Net and `automatic_penalty_params` is `False`. Specify the penalty parameter value for Elastic Net parameter 2.
`probability_entry`	number	Only available if the `method` selected is Forward Stepwise. Specify the significance level of the f statistic criterion for effect inclusion.
`probability_removal`	number	Only available if the `method` selected is Forward Stepwise. Specify the significance level of the f statistic criterion for effect removal.
`use_max_effects`	flag	Only available if the `method` selected is Forward Stepwise. Enables the `max_effects` control. When `False` the default number of effects included should equal the total number of effects supplied to the model, minus the intercept.
`max_effects`	integer	Specify the maximum number of effects when using the forward stepwise building method.
`use_max_steps`	flag	Enables the `max_steps` control. When `False` the default number of steps should equal three times the number of effects supplied to the model, excluding the intercept.
`max_steps`	integer	Specify the maximum number of steps to be taken when using the Forward Stepwise building `method`.
`use_model_name`	flag	Indicates whether to specify a custom name for the model (`true`) or to use the system-generated name (`false`). Default is `false`.
`model_name`	string	If `use_model_name` is `true`, specifies the model name to use.
`usePI`	flag	If `true`, predictor importance is calculated..
`perform_model_effect_tests`	boolean	Whether to perform model effect tests.
`non_neg_least_squares`	integer	Whether to perform non-negative least squares.

Was the topic helpful?

0/1000