Configuring model settings

The product uses default model settings to preprocess data and train models. However, you can configure your model settings to train your model. You can choose to automatically or manually configure the model settings. If you automatically configure the model settings, the product uses default values for model settings and data preprocessing. If you manually configure the model settings, you have more control of the values.

About this task

You can use the model settings function to set the configuration settings for currently selected asset type. You can switch asset types using the asset type menu on the asset panel.

When you upload your data, an initial training is automatically completed by using the following parameters:
  • Number of Trees: 3
  • Maximum Depth of Tree: 3
  • Learning Rate: 0.3
For the Gradient Boosting Decision Tree (GBDT) regression model type, the following parameters are available:
  • Number of Trees, range [2, 100], default 10
  • Maximum Depth of Tree, range [2, 10], default 4
  • Loss Function, default leastSquaresError, logLoss, and leastAbsoluteError
  • Learning Rate, range [0.01, 0.3], default 0.1

During training, the settings icon and Retrain button are disabled.

Procedure

  1. Click the settings icon.
  2. On the Model Settings tab, select the analysis interval to define how you want the data set to be aggregated before training. The default option for the analysis interval is daily or you can select weekly, monthly, yearly, or the raw option where no aggregation is done during training.
  3. In the Define Model Settings box, do one of the following options:
    1. Add the parameter values.
    2. Select Automatically define model parameters to have the values be automatically defined. If you select Automatically define model parameters, the training speed is slow.
  4. Optional: Select Automatically Retrain on Apply for the system to automatically start a training that is based on the new values.
  5. Select Apply.