AutoML forecasting: Split by model or grain

[pbartos]

I will use an example from your documentation:

https://docs.microsoft.com/en-us/azure/machine-learning/how-to-auto-train-forecast#preparing-data

What is the difference between train two models per store or one model with store grain? Will be the accuracy of one model with grains increased due to a larger amount of the data, etc?

[Enter feedback here]

---

Document Details

⚠ Do not edit this section. It is required for docs.microsoft.com ➟ GitHub issue linking.

• ID: bf6e8bb8-9f33-6315-dec3-915c82152f4f
• Version Independent ID: 0d184dff-a615-49e4-089a-8c41a8ff801c
• Content: Auto-train a time-series forecast model - Azure Machine Learning
• Content Source: articles/machine-learning/how-to-auto-train-forecast.md
• Service: machine-learning
• Sub-service: core
• GitHub Login: @nibaccam
• Microsoft Alias: nibaccam

[TitusA]

Building one large model will often be better than one model per series, but not always. If the series are quite heterogeneous (exhibit different patterns, span orders of magnitude), then it makes sense to fit multiple models. If they are similar, one model will generally do well.