Ensemble

class survivors.ensemble.base_ensemble.BaseEnsemble(size_sample=0.7, n_estimators=10, aggreg=True, ens_metric_name='roc', bootstrap=True, tolerance=True, aggreg_func='mean', **tree_kwargs)

Base ensemble of survival decision tree.

Attributes

size_samplefloat

Size of generated subsample

n_estimatorsint

Number of base models

ens_metric_namestr

Metric defines quantitative of ensemble

descend_metrboolean

Flag of descend for ens_metric_name

bootstrapboolean

Flag for using bootstrap sampling (with return)

toleranceboolean

Flag for fitting full ensemble and choosing best submodels

aggregboolean

Flag of aggregating base responses

aggreg_funcstr

Function of aggregating (if aggreg)

tree_kwargsdict

Parameters for building base models

modelslist

Base models of ensemble (for example, CRAID)

ooblist

Out of bag sample for each model

ens_metrarray-like

Values of ens_metric_name for ensemble with i models

list_pred_ooblist

Predictions of ensemble models[:i+1]

Methods

update_params : attributes preparation fit : build ensemble with X, y data (abstract) add_model : updating ensemble with new model and oob select_model : remaining fixed models tolerance_find_best : iterative method of selecting best sub ensemble

predict : return values of features, rules or schemes (look at CRAID) predict_at_times : return survival or hazard function

score_oob : calculate metric “ens_metric_name” for ensemble

class survivors.ensemble.base_ensemble.FastBaseEnsemble(size_sample=0.7, n_estimators=10, aggreg=True, ens_metric_name='roc', bootstrap=True, tolerance=True, aggreg_func='mean', **tree_kwargs)
class survivors.ensemble.bootstrap.BootstrapCRAID(**kwargs)

Bootstrap aggregation (Bagging) ensemble of survival decision tree. On each iteration probabilities of observations change by scheme.

Attributes

kwargsdict

Parameters for building base ensemble (look at BaseEnsemble)

Methods

fit : build ensemble with X, y data

class survivors.ensemble.bootstrap.ParallelBootstrapCRAID(**kwargs)

Parallel realization of BootstrapCRAID. Trees are built in parallel, and the number of processes is specified via n_jobs.

static fit_tree(x_sub, params)

An atomic out-of-class tree construction operation for parallelization.