Best Practices for ML Utterance Import

Machine Learning (ML) utterances play a crucial role in natural language understanding and bot performance.
Ensuring proper import processes is essential for maintaining optimal functionality and consistency across various environments.

This article provides detailed guidelines to help you navigate common issues associated with ML utterances import.

Importance of Proper Import

ML utterances are fundamental to the performance of bots, impacting their ability to understand and respond accurately to user inputs. Proper import processes ensure that these utterances function correctly, maintaining the bot’s overall performance and reliability.

Version Compatibility

Ensure Consistent Patch Versions

  • Consistent Patch Versions:
    Always ensure that both the source and target environments are running the same patch versions whenever possible. This practice reduces the risk of compatibility issues during the import process.

Upgrade Compatibility

  • Upgrading:
    While upgrading from an older version to a newer version (e.g., V10.1 to V10.2) is typically supported, downgrading from a newer version to an older version (e.g., V10.x to V9.x) is not recommended and should be avoided.

Unsupported Versions

  • Regular Checks:
    Regularly check for unsupported versions. If you encounter an unsupported version, inform the customer support (CS) team and request an upgrade. Additionally, you can negotiate with CS to waive off any Service Level Agreement (SLA) commitments related to tickets raised for unsupported versions.

Handling Import Issues

Manual Updates for Small Deltas

  • Discrepancies:
    If there are discrepancies between the source and target environments (e.g., a difference of 15-30 utterances), a practical workaround is to manually update the utterances in the target environment.
    This method helps to quickly address minor differences without major disruptions.

Exporting and Importing ML Only

  • Minimizing Issues:
    To minimize issues, consider exporting only the ML utterances and importing them into the target bot.
    This approach helps avoid complications related to patterns and other components, ensuring a smoother import process.

Avoid ML + Patterns Export/Import

  • Separate Exports:
    Refrain from exporting ML utterances along with patterns and importing them into the target bot. Combining these elements can lead to errors and inconsistencies, making it more challenging to maintain functionality.

Minimize Partial Imports

  • Complete Imports:
    As much as possible, avoid partial imports. Complete imports help maintain the integrity and consistency of the ML utterances across environments, reducing the likelihood of issues arising from incomplete data.