How to keep a deployed model effective| Intellipaat

Actions 1: Update the model with fresh data.

Get the reconstruction error value from the AE model using recent data when the model produced a high accuracy score and the negative test accuracy is also noticeably high. The amount of data change is minimal if the reconstruction error value is low. Retraining the model with recent data will be the best course of action after that.

Obtaining a Data Science Training is vital for upskilling and staying current in the workplace.

 

Actions 2:Retrain the model with new features.

The reconstruction error value is low, indicating that data hasn’t changed much, but the model needs to be tuned with new features when it produces low accuracy scores on both positive and negative datasets. Even though you already have the necessary data, retraining the same model on the same data won’t help in this situation. By performing the feature-engineering step, new features would be created, and the model would then be retrained with the new features. Keep in mind to keep the original model’s features.

 

Actions 3: Create a brand-new model from scratch.

When the model produced recent positive data with a low accuracy score and a negative test, the accuracy is noticeably high. Furthermore, the reconstruction error value for recent input data is also high, which is a definite sign that the new recent input data is very different from the model’s initial training data and has new features. Repetition of the feature extraction procedure would be the next course of action, followed by the creation and training of a new model from scratch.

Leave a Reply

Your email address will not be published. Required fields are marked *