e-Commerce Recommender Training Config

In your e-commerce scenario with frequently changing item menus, it's important to align your recommender system training with your data update frequency. While the auto-training happens every 7 days by default, this may not be optimal for your situation where the data changes every 4 days. Here are some suggestions to address this:

1. Manual retraining: Instead of relying on auto-training, you can manually retrain your models every 4 days to coincide with your data updates. This ensures that your recommendations are always based on the most current data.

2. Scheduled retraining: Set up a custom schedule to retrain your models every 4 days using AWS services like AWS Lambda and Amazon EventBridge. This automates the process while matching your data update frequency.

3. Monitor data changes: Implement a system to monitor significant changes in your data distribution. If substantial changes are detected, trigger a manual retraining of your models. This approach allows you to retrain only when necessary, potentially saving resources.

4. Hybrid approach: Use a combination of scheduled retraining every 4 days and monitoring for significant data changes. This ensures regular updates while also allowing for additional retraining if there are unexpected shifts in your data.

5. Adjust data import strategy: Consider importing a rolling window of data that extends beyond the 4-day menu cycle. This could provide more stability to your models while still capturing recent trends.

For the specific use cases of 'aws-ecomm-frequently-bought-together' and 'aws-ecomm-recommended-for-you', these approaches should work well. Remember that retraining models is a continuous process, especially in dynamic environments like e-commerce with frequently changing inventories. Regular retraining on newer data helps maintain the accuracy and relevance of your recommendations.

Sources
Retraining Models on New Data - Amazon Machine Learning