Minimum records needed for label to include it

Set a threshold for the minimum number of records a label must have in your dataset to be included in model training.

Avant de commencer

Verify that your dataset includes label values suitable for classification.

Role required: ml_admin or admin

Pourquoi et quand exécuter cette tâche

This advanced option helps to eliminate labels that occur infrequently. Setting this minimum threshold can reduce overfitting and unreliable predictions.

To configure, you provide a number for a minimum count of records. Any class value with fewer records than the minimum is dropped from the training set. Class values that are dropped before training can't appear in predictions.

To add this option using a script:

config.setMinLabelCount(50);

In JSON format:

{
    "min_label_count": 50
    }

Procédure

1. To add this option in the solution definition form, navigate to All > Predictive Intelligence > Classification > Solution Definitions.
2. Open an existing classification solution or create a new one.
3. Select the Advanced Solution Settings tab in the Related Links section of the form.
4. Select New to open the Advanced Solution Setting (ml_advanced_solution_settings) form.
5. In the Solution Parameters field, search for minimum records needed for label to include it.
6. In the User Inputs field, enter the number of records you want as the minimum, then select Submit.

   

Résultats

The setting appears as a row on the Advanced Solution Setting tab on your solution's form.

Que faire ensuite

Retrain your solution.

Concepts associés

• Using Machine Learning APIs

Tâches associées

• Create and train a classification solution