decForestCFit

Purpose

Fit a decision forest classification model.

Format

dfm = decForestCFit(y_train, x_train[, dfc])
Parameters:
  • y_train (Nx1 vector) – The target, or dependent variable.

  • X_train (NxP matrix) – The model features, or independent variables.

  • dfc (struct) –

    Optional input, an instance of the dfControl structure. For an instance named, dfc the members are:

    dfc.numTrees

    Scalar, number of trees (must be integer). Default = 100.

    dfc.pctObsPerTree

    Scalar, the percentage of observations selected for each tree (sampling with replacement). Valid range: 0.0 < pctObsPerTree <= 1.0. Default = 1.0.

    dfc.featuresPerSplit

    Scalar integer value, number of features considered as a possible split at each node. Default = sqrt(nvars).

    dfc.maxTreeDepth

    Scalar integer value, maximum tree depth. Default = 0 = unlimited.

    dfc.maxLeafNodes

    Scalar integer value, maximum number of leaves in each tree. Setting this to a positive integer value will cause the tree to be built by making the best possible splits first, instead of growing the trees in a depth first fashion. Default = 0 = unlimited.

    dfc.minObsLeaf

    Scalar integer value, minimum observations per leaf node. Default = 1.

    dfc.impurityThreshold

    Scalar, if the impurity value at a particular node is below this value, it will no longer be split. Default = 0.0.

    dfc.oobError

    Scalar, 1 to compute OOB error, 0 otherwise. Default = 0.

    dfc.variableImportanceMethod

    Scalar, method of calculating variable importance.

    • 0 = none,

    • 1 = mean decrease in impurity (Gini importance),

    • 2 = mean decrease in accuracy (MDA),

    • 3 = scaled MDA.

    Default = 0.

Returns:

dfm (struct) –

An instance of the dfModel structure. An instance named dfm will have the following members:

dfm.variableImportance

Matrix, 1 x p, variable importance measure if computation of variable importance is specified, zero otherwise.

dfm.oobError

Scalar, out-of-bag error if OOB error computation is specified, zero otherwise.

dfm.numClasses

Scalar, number of classes if classification model, zero otherwise.

Examples

new;
library gml;

rndseed 23423;

// Create file name with full path
fname = getGAUSSHome("pkgs/gml/examples/breastcancer.csv");

// Load all variables from dataset, except for 'ID'
data = loadd(fname, ". -ID");

// Separate dependent and independent variables
y = data[., "class"];
X = delcols(data, "class");

// Split data into 70% training and 30% test set
{ y_train, y_test, x_train, x_test } = trainTestSplit(y, X, 0.7);

// Declare 'df_mdl' to be an 'dfModel' structure
// to hold the trained model
struct dfModel df_mdl;

// Train the decision forest classifier with default settings
df_mdl = decForestCFit(y_train, X_train);

// Make predictions on the test set, from our trained model
y_hat = decForestPredict(df_mdl, X_test);

// Print classification quality report
call classificationMetrics(y_hat, y_test);

The code above will print the following output:

======================================================================
Model:              Decision Forest         Target variable:     class
Number Observations:            489         Number features:         9
Number of trees:                100           Obs. per Tree:      100%
Min. Obs. Per Node:               1     Impurity Threshhold:         0
======================================================================


========================================================================
Prediction Model:      DF Classification     Target variable:     class
Number Predictions:                  210     Number features:         9
========================================================================

===================================================
                             Classification metrics
===================================================
       Class   Precision  Recall  F1-score  Support

           0        0.99    0.99      0.99      155
           1        0.96    0.98      0.97       55

   Macro avg        0.98    0.98      0.98      210
Weighted avg        0.99    0.99      0.99      210

    Accuracy                          0.99      210

Remarks

The dfModel structure contains a fourth, internally used member, opaqueModel, which contains model details used by decForestPredict().

See also

Functions decForestPredict(), decForestRFit()