Spark Gradient Boosted Trees Learner (Regression)

Options

Settings

Target column

A numeric column that contains the values to train with. Rows with missing values in this column will be ignored during model training.

Feature Columns

The feature columns to learn the model with. Both nominal and numeric columns are supported. The dialog allows to select the columns manually (by moving them to the right panel) or via a wildcard/regex selection (all columns whose names match the wildcard/regex are used for learning). In case of manual selection, the behavior for new columns (i.e. that are not available at the time you configure the node) can be specified as either Enforce exclusion (new columns are excluded and therefore not used for learning) or Enforce inclusion (new columns are included and therefore used for learning).

Number of models

The number of Decision Tree models in the ensemble model. Increasing this number makes the model more expressive and improves training data accuracy. However, increasing it too much may lead to overfitting. Also, increasing this number directly increases the time required to train the ensemble, because the trees need to be trained sequentially.

Loss function

The loss function to minimize by the learner algorithm:

• squared (default): Squared Error, also known as L2 loss.
• absolute: Absolute Error, also known as L1 loss.

More information is available in the Spark documentation.

Max tree depth

Maximum depth of the Decision Trees. Must be >= 1.

Min rows per tree node

Minimum number of rows each tree node must have. If a split causes the left or right child node to have fewer rows, the split will be discarded as invalid. Must be >= 1.

Min information gain per split

Minimum information gain for a split to be considered.

Max number of bins

Number of bins to use when discretizing continuous features. Increasing the number of bins means that the algorithm will consider more split candidates and make more fine-grained decisions on how to split. However, it also increases the amount of computation and communication that needs to be performed and hence increases training time. Additionally, the number of bins must be at least the maximum number of distinct values for any nominal feature.

Advanced

Learning rate

Learning rate in interval (0, 1] for shrinking the contribution of each decision tree in the ensemble. This parameter should not need to be tuned often. Decreasing this value may improve stability, if the algorithm behavior seems unstable.

Data sampling (rows)

Sampling the rows is also known as bagging, a very popular ensemble learning strategy. If sampling is disabled (default), then each Decision Tree is trained on the full data set. Otherwise each tree is trained with a different data sample that contains the configured fraction of rows of the original data.

Feature sampling

Feature sampling is also called random subspace method or attribute bagging. Its most famous application are Random Forests, but it can also be used for Gradient Boosted Trees. This option specifies the sample size for each split at a tree node:

• Auto: If "Max number of models" is one, then this is the same as "All", otherwise "Square root" will be used.
• All (default): Each sample contains all features.
• Square root: Sample size is sqrt(number of features)
• Log2: Sample size is log2(number of features)
• One third: Sample size is 1/3 of the features

Use static random seed

Seed for generating random numbers. Randomness is used when sampling rows and features, as well as binning numeric features during splitting.

Views

This node has no views

Workflows

• No workflows found

Links

• No links available

Developers