Training a Model

Procedure

1. On the ExeML page, click the name of the project that has been created. The Label Data tab page is displayed. Select the label column and its data type.
2. On the Label Data tab page, click Train in the lower left corner. In the displayed Training Configuration dialog box, select an instance flavor used for training, click Next to go to the configuration page, confirm the specifications, and click Submit to start model training.

   For ExeML projects of predictive analytics, only the ExeML CPU (8U) instance flavor can be used for model training.

   The training takes a certain period of time. If you close or exit the page, the system continues training until it is complete.

3. On the Train Model tab page, wait until the training status changes from Running to Completed.
   Figure 1 Successful running
   
4. View the training details, such as the label column, data type, accuracy, and evaluation result.

   The example is a discrete value of binary classification. For details about the evaluation result parameters, see Table 1.

   For details about the evaluation results generated for different data types of label columns, see Evaluation Results.

An ExeML project supports multiple rounds of training, and each round generates a version. For example, the first training version is V001 (xxx), and the next version is V002 (xxx). The trained models can be managed by training version. After the trained model meets your requirements, deploy the model as a service.

Evaluation Results

The parameters in evaluation results vary depending on the training data type.

• Discrete values

  The evaluation parameters include recall, precision, accuracy, and F1 score, which are described in the following table.

  Table 1 Parameters in discrete value evaluation results

  Parameter	Description
  Recall	Fraction of correctly predicted samples over all samples predicted as a class. It shows the ability of a model to distinguish positive samples.
  Precision	Fraction of correctly predicted samples over all samples predicted as a class. It shows the ability of a model to distinguish negative samples.
  Accuracy	Fraction of correctly predicted samples over all samples. It shows the general ability of a model to recognize samples.
  F1 Score	Harmonic average of the precision and recall of a model. It is used to evaluate the quality of a model. A high F1 score indicates a good model.

• Continuous values

  The evaluation parameters include Mean Absolute Error (MAE), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE). The three error values represent a difference between a real value and a predicted value. During multiple rounds of modeling, a group of error values is generated for each round of modeling. Use these error values to determine the quality of a model. A smaller error value indicates a better model.