Package website: release | dev
Probabilistic Supervised Learning for mlr3.
What is mlr3proba ?
mlr3proba is a probabilistic supervised learning (PSL) toolkit for machine learning in R utilising the mlr3 package. Probabilistic supervised learning is a field of supervised machine learning in which probability distributions are predicted. Regression and classification tasks can be shown to be sub-fields of PSL, for example reducing a predicted probability distribution from probabilistic regression by taking its mean gives a regression prediction. PSL is therefore a powerful tool to provide more information about predictions than the more classical regression or classification. Probably the most known variant of PSL is survival analysis, where the task of interest is to predict an individual’s survival curve. Other forms of PSL include density estimation and probabilistic regression. To-date, PSL toolkits in R have been limited to Bayesian simulation packages, but mlr3proba hopes to change this by allowing domain-agnostic (Bayesian or Frequentist) fit/predict and evaluation workflows.
Survival Analysis
Survival Learners
| ID | Learner | Package |
|---|---|---|
| surv.blackboost | Gradient Boosting with Regression Trees | mboost |
| surv.coxph | Cox Proportional Hazards | survival |
| surv.cvglmnet | Cross-Validated GLM with Elastic Net Regularization | glmnet |
| surv.flexible | Flexible Parametric Spline Models | flexsurv |
| surv.gamboost | Gradient Boosting for Additive Models | mboost |
| surv.gbm | Generalized Boosting Regression Modeling | gbm |
| surv.glmboost | Gradient Boosting with Component-wise Linear Models | mboost |
| surv.glmnet | GLM with Elastic Net Regularization | glmnet |
| surv.kaplan | Kaplan-Meier Estimator | survival |
| surv.mboost | Gradient Boosting for Generalized Additive Models | mboost |
| surv.nelson | Nelson-Aalen Estimator | survival |
| surv.parametric | Fully Parametric Survival Models | survival |
| surv.penalized | L1 and L2 Penalized Estimation in GLMs | penalized |
| surv.randomForestSRC | RandomForestSRC Survival Forest | randomForestSRC |
| surv.ranger | Ranger Survival Forest | ranger |
| surv.rpart | Rpart Survival Forest | rpart |
| surv.svm | Regression, Ranking and Hybrid Support Vector Machines | survivalsvm |
Survival Measures
| ID | Learner | Package |
|---|---|---|
| surv.beggC | Begg’s C-Index | survAUC |
| surv.chamblessAUC | Chambless and Diao’s AUC | survAUC |
| surv.gonenC | Gonen and Heller’s C-Index | survAUC |
| surv.graf | Integrated Graf Score | mlr3proba |
| surv.grafSE | Standard Error of Integrated Graf Score | mlr3proba |
| surv.harrellC | Harrell’s C-Index | mlr3proba |
| surv.hungAUC | Hung and Chiang’s AUC | survAUC |
| surv.intlogloss | Integrated Log Loss | mlr3proba |
| surv.intloglossSE | Standard Error of Integrated Log Loss | mlr3proba |
| surv.logloss | Log Loss | mlr3proba |
| surv.loglossSE | Standard Error of Log Loss | mlr3proba |
| surv.nagelkR2 | Nagelkerke’s R2 | survAUC |
| surv.oquigleyR2 | O’Quigley, Xu, and Stare’s R2 | survAUC |
| surv.songAUC | Song and Zhou’s AUC | survAUC |
| surv.songTNR | Song and Zhou’s TNR | survAUC |
| surv.songTPR | Song and Zhou’s TPR | survAUC |
| surv.unoAUC | Uno’s AUC | survAUC |
| surv.unoC | Uno’s C-Index | survAUC |
| surv.unoTNR | Uno’s TNR | survAUC |
| surv.unoTPR | Uno’s TPR | survAUC |
| surv.xuR2 | Xu and O’Quigley’s R2 | survAUC |
Feature Overview and Lifecycle
The vision of mlr3proba is to be the first complete probabilistic machine learning package in R. This encompasses survival analysis, probabilistic regression, and unsupervised density estimation. The first release of mlr3proba is focused entirely on survival analysis and introduces TaskSurv. Later releases will include TaskDensity and will extend TaskRegr to have probabilistic predict types. The lifecycle of the survival task and features are considered maturing and any major changes are unlikely. The density and probabilistic regression tasks are currently in the early stages of development. The current main features of mlr3proba are:
- The added
TaskSurv,LearnerSurv,PredictionSurvfor survival analysis - 17 survival learners, and 21 survival measures, including efficient implementations of censoring-adjusted probabilistic measures, such as the Integrated Graf (or Brier) Score.
- PipeOps integrated with mlr3pipelines for composition of probability distributions from linear predictors
Future Plans
- Add
TaskDensity,PredictionDensity,LearnerDensity, and associated learners/measures - Add
probpredict type toTaskRegr, and associated learners/measures - Allow
MeasureSurvto return measures at multiple time-points simultaneously - Improve estimation of integrated scores, and re-implement survAUC scores in mlr3proba
- Continue to add survival measures and learners
Bugs, Questions, Feedback
mlr3proba is a free and open source software project that encourages participation and feedback. If you have any issues, questions, suggestions or feedback, please do not hesitate to open an “issue” about it on the GitHub page!
In case of problems / bugs, it is often helpful if you provide a “minimum working example” that showcases the behaviour (but don’t worry about this if the bug is obvious).
Please understand that the resources of the project are limited: response may sometimes be delayed by a few days, and some feature suggestions may be rejected if they are deemed too tangential to the vision behind the project.
Similar Projects
A predecessor to this package is mlr, using the survival task. Several packages exist for pure Bayesian probabilistic modelling, including jags and stan. For implementation of a few survival models and measures, the largest package is survival. There does not appear to be a package that implements many different variants of density estimation, but see this list for the biggest density estimation packages in R.
