NEWS.md
pmml() now has a model_version parameter that can be used to add a modelVersion attribute to the PMML Header.xform_function() now supports "numeric" and "factor" new_field_data_type.save_pmml() no longer has the version parameter, since r-pmml exports PMML compliant with only the latest schema.pmml.ARIMA() no longer exports models with the Exact Least Squares method. Exports where ts_type = "arima" produce PMML with Conditional Least Squares representation only. The deprecated exact_least_squares parameter has been removed.pmml.ARIMA() now exports models with StateSpaceModel representation by default.xform_function() no longer creates an unnecessary "NA" level in field_data$dataType.xform_wrap() no longer sets field_data$dataType to factor for all columns of a tibble.pmml.xgb.Booster() no longer switches probabilities for "binary:logistic" objective function.add_data_field_attributes() now throws appropriate errors instead of using print() statements.pmml.ARIMA() can now export models with StateSpaceModel representation. This is controlled by a new parameter, ts_type. The default option (ts_type = "arima") still exports an ARIMA representation as before. When ts_type = "statespace", the exported PMML uses a StateSpaceModel element.exact_least_squares parameter in pmml.ARIMA() is now deprecated. In a future release, all exports where ts_type = "arima" will be exported with the Exact Least Squares method.pmml.iForest: anomalyThreshold -> anomaly_threshold
pmml.kmeans: algorithm.name -> algorithm_name
rename_wrap_var: wrap_data -> wrap_object
xform_norm_discrete: inputVar -> input_var
pmml.ARIMA() now exports prediction intervals for non-seasonal models.pmml.ARIMA() can now export seasonal models with Exact Least Squares representation, and does so by default.Package now exports PMML with schema version 4.4.
pmml.ARIMA() function added - exports ARIMA time series models from the forecast package.
pmml.svm() now has a detect_anomaly argument, allowing the user to specify whether the PMML detects anomalies or inliers. The exported PMML now has two OutputField elements: anomalyScore and one of anomaly or inlier.
The following arguments are deprecated. They can still be used, but will produce a warning message and will be removed in a future release.
pmml.iForest: anomalyThreshold -> anomaly_threshold
pmml.kmeans: algorithm.name -> algorithm_name
rename_wrap_var: wrap_data -> wrap_object
xform_norm_discrete: inputVar -> input_var
Application version in PMML Header corresponds to pmml package version.
pmml.iForest now uses attribute sampleDataSize instead of element ParameterList to store the model$phi value.
pmml.gbm now adds modelName attribute to the final segment for multinomial gbm models.
testthat file names correspond to the functions being tested.
Edited make_output_nodes doc for clarity.
Updated formatting in vignettes.
Fixed spelling and added word list via spelling package.
With this release, pmmlTransformations has been merged into pmml and package development has been moved to GitHub. This was also a good opportunity to apply a style and rename many functions and parameters to make the code more uniform and easier to understand.
We used the tidyverse style guide when renaming functions, parameters, and arguments. In addition, some parameters and arguments have been renamed for clarity, and several functions were removed.
For functions that use a dots (...) parameter, the old parameters will still be accepted, even though these old parameters will not be used.
pmml() parameters and default arguments have been changed as follows:
model.name = "Rattle_Model" -> model_name = "R_Model"
app.name = "Rattle/PMML" -> app_name = "SoftwareAG PMML Generator"
unknownValue -> missing_value_replacement
pmml.iForest()
parentInvalidValueTreatment -> parent_invalid_value_treatment
childInvalidValueTreatment -> child_invalid_value_treatment
pmml.lm()
dataset argument has been removed.pmml.naiveBayes()
predictedField - predicted_field
pmml.randomForest()
unknownValue -> missing_value_replacement
parentInvalidValueTreatment -> parent_invalid_value_treatment
childInvalidValueTreatment -> child_invalid_value_treatment
pmml.xgb.Booster()
inputFeatureNames -> input_feature_names
outputLabelName -> output_label_name
outputCategories -> output_categories
xgbDumpFile -> xgb_dump_file
parentInvalidValueTreatment -> parent_invalid_value_treatment
childInvalidValueTreatment -> child_invalid_value_treatment
AddAttributes() -> add_attributes()
xmlmodel -> xml_model
addDDAttributes() -> add_data_field_attributes()
addDFChildren() -> add_data_field_children()
addMSAttributes() -> add_mining_field_attributes() xmlmodel -> xml_model
addOutputField() -> add_output_field()
xmlmodel -> xml_model
makeIntervals() -> make_intervals()
makeOutputNodes() -> make_output_nodes()
makeValues() -> make_values()
pmmlTransformations have been merged into pmml and had the following name and parameter changes:
xformInfo -> xform_info
boxdata -> wrap_object (except in RenameVar)mapMissingTo -> map_missing_to
DiscretizeXform() -> xform_discretize()
defaultValue -> default_value
mapMissingTo -> map_missing_to
FunctionXform() -> xform_function()
origFieldName -> orig_field_name
newFieldName -> new_field_name
newFieldDataType -> new_field_data_type
formulaText -> expression
MapXform() -> xform_map()
defaultValue -> default_value
MinMaxXform() -> xform_min_max()
NormDiscreteXform() -> xform_norm_discrete()
RenameVar() -> rename_wrap_var()
boxdata -> wrap_data
WrapData() -> xform_wrap()
indata -> data
useMatrix -> use_matrix
ZScoreXform() -> xform_z_score()
Add support for one-class svm (anomaly detection) models from e1071
Add support for iForest (anomaly detection) models from isofor
Add support for boolean values in functionToPMML
Fix missing nodes issue in xgboost caused by deep trees
Update savePMML to output LF line endings, independent of OS
Update Description and pmml.R documentation
Add dataType argument to every Output node
Fix the family check in pmml.cv.glmnet to only allow “gaussian” and “poisson”
Add note about supported family types in pmml.cv.glmnet documentation
Add note about sparse matrices in pmml.xgb.Booster documentation
Add vignette with list of supported packages and models
Fix bug in xgboost when using transformations
Add invalidValueTreatment=“returnInvalid” in MiningSchema by default for all models
Add arguments for invalidValueTreatment attributes to pmml.xgboost() and pmml.randomForest()
Fix bug due to 1- vs 0-based indexing in gbm converter
Change default pmml version to 4.3 in gbm converter
Fix comment added at top of XML doc by savePMML()
Fix bug in xgboost regression end element
Fix bug in xgboost when feature names are substrings of each other
Fix bug in xgboost binomial model
Add check for xgboost objectives
Add support for multi:softmax objective in xgboost
Update pmml.rfsrc.R doc to state dependency on version 2.5.0 or earlier
Add converter for xgboost models
Add converter for gbm models
Add converter for neighbr model
Output error on detecting an unsupported kernel in kernlab::ksvm() models
Add examples and note to pmml.glm.Rd
Output error on detecting unsupported feature classes
Output error on detecting interaction terms in linear models
Fix error created when fitting random forest model with 1 tree
Update references to PMML spec 4.3
Add functionToPMML
Add addDDInfo
Add addOutputInfo
Edit to work with pmmlTransformations::FunctionXform: pmml, pmml.datadictionary and pmml.miningschema
Add warning message to addLT; will deprecate in next version
Add support for the svm model in e1071 package
Add support for ada package
Add classificationMethod attribute for ksvm model (kernlab) for information purpose
Add usageType for AssociationModel MiningField elements
Fix bug for combining algorithm PMML and transformation XML fragment
Remove obsolete codes
Updates to pmml.arules from Michael Hahsler.
Begin using a ChangeLog (this) file and simply dump previous changes below.
Begin support for PMML 4.0.
Small changes to pmml.lm to support UTF-8/Japanese.
Fix spelling error in DESCRIPTION - CRAN check found it.
Various updates from Zementis to the randomForest schema. All the mining schemas in each segment of the random forest now also output the attribute invalidValueTreatment as asis.
Change standard to have an R file per exported function, and all internal functions begin with .
– Graham Williams Graham.Williams@togaware.com Sun, 09 Dec 2012 22:28:00 +1100