Training a Model

Procedure

1. On the ExeML page of the new version, click the name of the target project to view the execution status of the current workflow.
2. On the predictive analytics phase, wait until the phase status changes from Running to Executed.
3. Click to 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 an AI application version. For example, the first training version is 0.0.1, and the next version is 0.0.2. 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.