| Day | Focus | Start Date | Completion Date | Reference |
|---|---|---|---|---|
| 2 | Initial Model Setup: - Configure the algorithm framework - Load initial datasets for processing - Start preliminary training routines | 10/11/2025 | 11/11/2025 | AWS Personalize Docs https://docs.aws.amazon.com/personalize/ |
| 3 | Review early outputs and investigate anomalies or unexpected behaviors | 12/11/2025 | 12/11/2025 | |
| 4 | Data Refinement: - Introduce additional data fields - Clean and reconcile dataset inconsistencies - Repeat training cycle with revised data | 12/11/2025 | 13/11/2025 | |
| 5 | Assess updated model performance and examine metric shifts | 14/11/2025 | 14/11/2025 | |
| 6 | Iterative observations and planning next refinement steps |
This week highlighted the practical challenges of building a recommendation system and working with constrained data environments:
Limited Interaction Signals:
The dataset mainly tracks completed actions (“Bookings”), lacking intermediary signals like “Views” or “Clicks”. This scarcity constrains the system’s ability to infer deeper user preferences, reflected in modest evaluation scores.
Validation Constraints:
Without a fully functional front-end interface, real-world validation remains largely theoretical. Current assessment relies on quantitative metrics rather than intuitive, visual confirmation.
Pipeline Sensitivity to Schema Changes:
Introducing new data attributes requires revisiting the preprocessing pipeline. Each adjustment involves re-cleaning, re-mapping, and retraining, highlighting the delicacy of the end-to-end workflow.
Key Takeaways:
While specific cloud tools support dataset handling and model training, the main insights are procedural: ensuring dataset completeness, maintaining flexible pipelines, and carefully managing iterative training cycles.