simRegression Model Specifications

Model Specifications modify a model’s behavior and access to server resources.

simRegression parameters guide the model training process by dictating the number and size of decision trees to be constructed, as well as other aspects of the learning procedure. [Therefore, it is highly recommended to use a grid to select the appropriate parameters. <- We should add this once we support grids for Regression.]

Iterations	Number of gradient boosting algorithm iterations/trees. More iterations can improve accuracy but increase overfitting risk.
Max Depth	Maximum tree depth. Deeper trees capture more complex patterns but may overfit.
ETA	Learning rate shrinkage to prevent overfitting. Smaller ETA values slow down learning and help in avoiding overfitting.
Col Sample by Tree	Feature sub-sample ratio per tree. Lower ratios increase diversity and prevent overfitting.
Regression Objective	Loss function used for training. It should match the regression problem's specifics. The allowed values are: squared_error: Minimizes the squared difference between actual and predicted values, emphasizing larger errors. squared_log_error: Uses squared logarithmic loss, suitable for targets greater than -1. logistic: Utilized in logistic regression, it outputs probabilities. pseudo_huber_error: Employs Pseudo Huber loss, an alternative to absolute loss. absolute_error: Involves L1 error for regression.

simRegression Model Specifications

Was this article helpful?

Comments

<%= previousTitle %>

<%= nextTitle %>

Table of Contents

<%= heading %>

<%= block.name %>

<%= heading %>

Categories

Toggle navigation menu

<%= category.name %>