Models¶

UncoverML has a variety of algorithms to choose from.

Config shows the algorithm parameter to add to the learning block of the config to use the model.

The majority of models are subclassed from models available in scikit-learn or the NICTA Revrand package. Follow the links under Documentation and parameters to view the specific model documentation and available parameters from these resources.

Additional parameters covers parameters specific to UncoverML that can be added to the arguments section of the learning block.

Regressors¶

Random Forest¶

Config:

algorithm: randomforest

Documentation and parameters:

Additional parameters:

target_transform

Multi Random Forest¶

An ensemble of Random Forest predictors.

Config:

algorithm: multirandomforest

Documentation and parameters:

Additional parameters:

target_transform
forests: number of Random Forest submodels
parallel: boolean, whether to train this model in parallel
- If parallel is True, this model can be trained using multiple processors. See Multiprocessing and Partitioning.

Bayes Regression/Standard Linear Model¶

Config:

algorithm: bayesreg

Documentation and parameters:

Additional parameters:

target_transform

Stochastic Gradient Descent Bayes Regression/Standard Linear Model¶

Config:

algorithm: sgdbayesreg

Documentation and parameters:

Additional parameters:

target_transform

Approximate Gaussian Process¶

Subclass of Bayes Regression/Standard Linear Model

Config:

algorithm: approxgp

Documentation and parameters:

Additional parameters:

target_transform

Stochastic Gradient Descent Approximate Gaussian Process¶

Subclass of Stochastic Gradient Descent Bayes Regression/Standard Linear Model

Config:

algorithm: sgdapproxgp

Documentation and parameters:

Additional parameters:

target_transform

Support Vector Regression¶

Config:

algorithm: svr

Documentation and parameters:

Additional parameters:

target_transform

Automatic Relevance Determination Regression¶

Config:

algorithm: ardregression

Documentation and parameters:

Additional parameters:

target_transform

Decision Tree¶

Config:

algorithm: decisiontree

Documentation and parameters:

Additional parameters:

target_transform

Extra Tree¶

Config:

algorithm: extratree

Documentation and parameters:

Additional parameters:

target_transform

Cubist¶

Config:

algorithm: cubist

Documentation and parameters:

Additional parameters:

target_transform

Multi Cubist¶

Config:

algorithm: multicubist

Documentation and parameters:

Additional parameters:

target_transform
trees: number of Cubist submodels to train
parallel: boolean, whether to train this model in parallel
- If parallel is True, this model can be trained using multiple processors. See Multiprocessing and Partitioning.

K Nearest Neighbour¶

Config:

algorithm: nnr

Documentation and parameters:

Additional parameters:

target_transform

Bootstrapped SVR¶

Allows probabilistic predictions for SVR by taking statistics from an ensemble of SVR models predicting on bootstrapped (resampled) data.

Config:

algorithm: bootstrapsvr

Documentation and parameters:

Additional parameters:

target_transform
n_models: int, number of models to train (i.e. number of times to resample data)
parallel: boolean, whether to train this model in parallel
- If parallel is True, this model can be trained using multiple processors. See Multiprocessing and Partitioning.

Random Forest (Optimisable)¶

This is a subclass of Random Forest structured to be compatible with optimisation.

Config:

algorithm: transformedrandomforest

Documentation and parameters:

Additional parameters:

target_transform

Gradient Boost (Optimisable)¶

Config:

algorithm: gradientboost

Documentation and parameters:

Additional parameters:

target_transform

Gaussian Process Regressor (Optimisable)¶

Config:

algorithm: transformedgp

Documentation and parameters:

Additional parameters:

target_transform

Stochastic Gradient Descent Regressor (Optimisable)¶

Config:

algorithm: sgdregressor

Documentation and parameters:

Additional parameters:

target_transform

Support Vector Regression (Optimisable)¶

Duplicate of Support Vector Regression structured to be compatible with optimisation.

Config:

algorithm: transformedsvr

Documentation and parameters:

Additional parameters:

target_transform

Ordinary Least Squares Regression (Optimisable)¶

Subclass of base scikit-learn LinearRegression.

Config:

algorithm: ols

Documentation and parameters:

uncoverml.optimise.models.TransformedOLS

Additional parameters:

target_transform

Elastic Net (Optimisable)¶

Config:

algorithm: elasticnet

Documentation and parameters:

Additional parameters:

target_transform

Huber (Optimisable)¶

Config:

algorithm: huber

Documentation and parameters:

Additional parameters:

target_transform

XGBoost (Optimisable)¶

Config:

algorithm: xgboost

Documentation and parameters:

Additional parameters:

target_transform

Interpolators¶

Linear ND Interpolator¶

Config:

algorithm: linear

Documentation and parameters:

Nearest ND Interpolator¶

Config:

algorithm: nn

Documentation and parameters:

RBF Interpolator¶

Config:

algorithm: rbf

Documentation and parameters:

CT Interpolator¶

Config:

algorithm: cubic2d

Documentation and parameters:

Classifiers¶

Label encoding is performed implicitly on UncoverML classifiers.

Logistic Classifier¶

Config:

algorithm: logistic

Documentation and parameters:

Logistic RBF Classifier¶

Kernelized version of LogisticClassifier

Config:

algorithm: logisticrbf

Documentation and parameters:

Random Forest Classifier¶

Config:

algorithm: forestclassifier

Documentation and parameters:

Support Vector Classifier¶

Config:

algorithm: svc

Documentation and parameters:

Boosted Trees¶

Config:

algorithm: boostedtrees

Documentation and parameters: