Skip to content

MedPredictor practical guide

  • Linear model : type = "linear_model" - linear regression using coordinate descent with Ridge factors
  • Linear-SGD : type = "linear_sgd" - linear regression using stochastic gradient descent
  • GDLM : type = "gdlm" - Generalized linear model using stochastic gradient descent
  • QRF : type = "qrf" - Random Forest using quantization for efficency
  • KNN : type = "knn" - K Nearest Neighbors
  • MARS : type = "mars" - (quite slow) impliminationg of MARS (Multivariate Additive Regression Spline)
  • GBM : type = "gbm" - Gradient Boosting Machine  - imported public library used in R
  • BP : type = "BP" - Back propogation. (Obsolete) version of NN
  • Multi-Class : type = "multi_class" - an enevelope for multi-class problems, currently implementing one-vs-all
  • XGBoost : type = "xgb" - public implementation of xgboost + internaly added features
  • Lasso : type = "lasso" - Lasso linear regression using gradient descent
  • MicNet : type = "micNet" - Neural Net
  • Booster : type = "booster" - Envelope for boosting models
  • Deep-Bit : type = "deep_bit" - Nir's deep bit using Yoav's implementation
  • Light-GBM : type = "light_gbm" - public implementation of fast GBM
  • SVM : type = "svm" - Support Vector Machine
  • multi-models : type = "multi_models" - an envelope for multiple models for different inputs (e.g. - age specific models)
  • vw : type = "vw" - an envelope for Vowpal Wabbit models
  • TQRF : type = "tqrf" - new version of quantized RF (Not ready yet ?)
  • BART : type = "bart" - public implementation of BART (Bayesian Additive Regression Trees)
  • MASK: type="by_missing_value_subset" - see here