View the model evaluation
Reviewing model evaluation is an important step in developing a custom classification model. It helps you learn how well your model is performing, and gives you an idea about how it will perform when used in production.
Prerequisites
Before you train your model you need:
- A custom classification project with a configured Azure blob storage account,
- Text data that has been uploaded to your storage account.
- Tagged data
- A successfully trained model
See the application development lifecycle for more information.
Model evaluation
The evaluation process uses the trained model to predict user-defined classes for files in the test set, and compares them with the provided data tags. The test set consists of data that was not introduced to the model during the training process.
View the model details using Language Studio
Go to your project page in Language Studio.
Select View model details from the left side menu.
In this page you can only view the successfully trained models. You can select the model name for more details.
You can find the model-level evaluation metrics under the Overview section and the class-level evaluation metrics under the Class performance metrics section. See Evaluation metrics for more information.
Note
If you don't find all the classes displayed here, it is because there were no tagged files of this class in the test set.
Under the Test set confusion matrix, you can find the confusion matrix for the model.
Note
The confusion matrix is currently not supported for multiple label classification projects.
Single label classification
Next steps
As you review your how your model performs, learn about the evaluation metrics that are used. Once you know whether your model performance needs to improve, you can begin improving the model.