Skip to main content

Backtesting and its importance

This deep dive explains the importance of backtesting.

Learning Objectives​

By the end of this deep dive, you will understand:

  • what is backtesting
  • how to backtest a trained a recipe
  • how to improve a recipe using backtesting

What is Backtesting?​

  • Backtesting is a method used to test new changes or settings on a trained recipe using previous images, especially those it didn’t perform well on.
  • It helps evaluate how the updated recipe would perform on known data before applying it to new cases.
  • You can backtest images on these settings:
    • Template Image and alignment
    • Inspection Setup
    • Classification Block

Settings that can be backtested

Backtesting on Template Image and Alignment​

You can upload the failed capture to adjust the recipe settings and ensure it adapts correctly to the change.

Testing the Template Image and Alignment Block

The variation between the aligned reference image and the detected image highlights misalignment, which is corrected through this process.

Trained alignerFound alignerFixed aligner and backtested
Example of trained alignerExample of found alignerExample of fixed aligner and backtested

Backtesting on Inspection Setup​

You can upload the failed capture to adjust the ROI settings and ensure it adapts correctly to the change.

Testing the Inspection Block

Backtesting on Classification Block​

Upload the failed image to check if the model’s performance has improved after adjustments or retraining.

Testing the Classification Block