FastTreeRegressionTrainer.Options Class

Create a new FastTreeRegressionTrainer.Options object with default values.

When a root split is impossible, allow training to proceed.

Percentage of training examples used in each bag. Default is 0.7 (70%).

Number of trees in each bag (0 for disabling bagging).

Option for using best regression step trees.

Bias for calculating gradient for each feature bin for a categorical feature.

Bundle low population bins. Bundle.None(0): no bundling, Bundle.AggregateLowPopulation(1): Bundle low population, Bundle.Adjacent(2): Neighbor low population bundle.

Whether to do split based on multiple categorical feature values.

Compress the tree Ensemble.

Whether to utilize the disk or the data's native transposition facilities (where applicable) when performing the transpose.

Dropout rate for tree regularization.

Enable post-training tree pruning to avoid overfitting. It requires a validation set.

The entropy (regularization) coefficient between 0 and 1.

Column to use for example weight.

Print execution time breakdown to ML.NET channel.

Column to use for features.

The feature first use penalty coefficient.

Whether to collectivize features during dataset preparation to speed up training.

The fraction of features (chosen randomly) to use on each iteration. Use 0.9 if only 90% of features is needed. Lower numbers help reduce over-fitting.

The fraction of features (chosen randomly) to use on each split. If it's value is 0.9, 90% of all features would be dropped in expectation.

The feature re-use penalty (regularization) coefficient.

The seed of the active feature selection.

Filter zero lambdas during training.

Tree fitting gain confidence requirement. Only consider a gain if its likelihood versus a random choice gain is above this value.

Sample each query 1 in k times in the GetDerivatives function.

The number of histograms in the pool (between 2 and numLeaves).

Column to use for labels.

The learning rate.

Maximum number of distinct values (bins) per feature.

Maximum categorical split groups to consider when splitting on a categorical feature. Split groups are a collection of split points. This is used to reduce overfitting when there many categorical features.

Maximum categorical split points to consider when splitting on a categorical feature.

Number of post-bracket line search steps.

Upper bound on absolute value of single tree output.

Print memory statistics to ML.NET channel.

The minimal number of data points required to form a new tree leaf.

Minimum categorical example percentage in a bin to consider for a split. Default is 0.1% of all training examples.

Minimum categorical example count in a bin to consider for a split.

Minimum line search step size.

The max number of leaves in each regression tree.

The number of threads to use.

Total number of decision trees to create in the ensemble.

Optimization algorithm to be used.

The tolerance threshold for pruning.

The moving window size for pruning.

Training starts from random ordering (determined by /r1).

Column to use for example groupId.

The seed of the random number generator.

Shrinkage.

Smoothing parameter for tree regularization.

The temperature of the randomized softmax distribution for choosing the feature.

Sparsity level needed to use sparse feature representation.

Calculate metric values for train/valid/test every k rounds.

Determines whether to use line search for a step size.

Use window and tolerance for pruning.

Write the last ensemble instead of the one determined by early stopping.

Early stopping metrics.

Early stopping rule used to terminate training process once meeting a specified criterion. Possible choices are EarlyStoppingRuleBase's implementations such as TolerantEarlyStoppingRule and GeneralityLossRule.