Train ML Model#

MLModel.Edit

Overview#

This operation is used to train an ML Model.

Setup#

No specific setup is required other than to meet the preconditions of the transaction.

Preconditions#

• The ML Model exists in the system and a transformation has been applied.

Sequence of Steps#

When the ML Model has been created, you can proceed with training.

To conclude the training effectively, the ML Model goes through a scaling process.

Transformations#

1. On the top menu, select Transform.
2. Choose a field to predict by marking it as a Label.
3. Optionally, you may enable cleaning of null values or removing outliers.
4. Optionally, if you do not select a field as a Label, the ML Model will automatically be set to Unsupervised.
5. Select the percentage of data that should be used to train/validate/test the model.
6. Transform the data.
7. After this scaling operation is successfully completed, the ML Model is now ready to train.

When the transformation is applied, the normalized features can be observed in the Transformed Data section.

Training#

1. Select Train on the top menu to train the model by choosing how long the AutoML algorithm should run and the metric to optimize.

   

2. AutoML will automatically train numerous models using different machine learning algorithms with varying hyper-parameters, and then return the best model according to the specified metric. You may also define a timeout for training. If there is no significant improvement on the validation set and the ML Model accuracy starts to steadily decrease, the ML Model will stop training or it will continue to train until it reaches the timeout you defined.

   

   Info

   In case of unsupervised learning, the model will ask you to define: the number of estimators (trees), the layer depth, and the number of samples.

3. When trained, the model can be tested to evaluate its accuracy.

   

4. Finally, the model can be deployed so that it can start outputting predictions in real time.