Spark Logistic Regression Learner

Options

Regularizer

The purpose of the regularizer is to encourage simple models and avoid overfitting. For more details on supported regularizers see the Regularizers section of the MLlib documentation.

Regularization

The fixed regularization parameter r >= 0 defines the trade-off between the two goals of minimizing the loss (i.e., training error) and minimizing model complexity (i.e., to avoid overfitting).

Number of iterations

The number of iterations the method should run.

Optimization Method

Under the hood, linear methods use convex optimization methods to optimize the objective functions. MLlib uses two methods, Stoachastic Gradient Descent (SGD) and Limited-memory BFGS (L-BFGS) which are described in detail in the MLlib optimization section.

Number of corrections

The number of corrections used in the L-BFGS update. Only available for L-BFGS.

Tolerance

The convergence tolerance. The tolerance is a condition which decides iteration termination. The end of iteration is decided based on below logic.

• If the norm of the new solution vector is > 1, the diff of solution vectors is compared to relative tolerance which means normalizing by the norm of the new solution vector.
• If the norm of the new solution vector is <= 1, the diff of solution vectors is compared to absolute tolerance which is not normalizing.

Only available for L-BFGS.

Loss function

For more details on the supported loss functions see the Loss function section of the MLlib documentation.

Step size

The initial step size of SGD for the first step. In subsequent steps, the step size will decrease with stepSize/sqrt(t). Only available for SGD.

Fraction

The fraction of data to be used for each SGD iteration. The default of 1.0 corresponds to deterministic/classical gradient descent. Only available for SGD.

Use feature scaling

Select this option to use feature scaling before model training to reduce the condition numbers which can significantly help the optimizer converging faster. Whether to perform feature scaling before model training to reduce the condition numbers which can significantly help the optimizer converging faster. The scaling correction will be translated back to resulting model weights, so it's transparent to users.

Add intercept

Select this option to add intercept.

Validate data

Select this option if the algorithm should validate data before training.

Class column

The classification column. Must be numeric.

Feature Columns

The feature columns to learn the model from. Supports only numeric columns.

Views

This node has no views

Workflows

• No workflows found

Links

• No links available

Developers