Does PlanIQ retrain the forecast model everytime I run the corresponding forecast action?

msburdak

I have a time-series forecasting use-case. I am forecasting on a monthly basis, i.e. the forecast horizon is 31 days. As I move into future I'll have more actuals data available, and I want to include those actuals also in the training. For example, while predicting for Jan 24 the training set was upto Dec 24, now i want to predict for Feb 24 and the training set needs to be upto Jan 24.

Do I need to create a new forecast model? Or would updating my data collection and running the forecast action only would ensure that model uses the new training set?

Anurag0911

PlanIQ retains the model structure and parameters from previous training sessions. This means that the model itself is not discarded after each forecast. When you run the forecast action with updated data, PlanIQ does not automatically retrain the model from scratch.

However, without retraining, the model will not have learned from the new actuals data.

To ensure that your forecast model uses the new training set with updated actuals, you need to:

1. Update your data collection with the latest actuals.
2. Retrain the model with the updated data.
3. Run the forecast action to generate the new forecast.

I hope this answers your question

seymatas1

You have data collections, forecast models, and forecast actions in PlanIQ.

Data Collections

• To train your models with the latest data, create a new data collection. You can do this by duplicating the last data collection and giving it an appropriate name and import action, for example, "PlanIQ Data Collection Sales Forecasting August 2024."
• You don't have to train your models every month; you can choose to train them quarterly or at other intervals. If you decide not to train them monthly, use the “Rerun Data Assembly” option instead.

However, since you have limited months of data, I recommend training your models every month if you have the resources to perform these manual steps regularly.

Forecast Models

• If you decide to train your models every month, create a new data collection and update the forecast model name and the associated data collection.
• If you didn’t create a new data collection, you can skip this step.
  

Forecast Actions

• If you retrained your models with the latest data and created new forecast models, update your existing forecast actions to use the newly created models instead of the previous month’s models.

• If you skipped the retraining step, simply run your existing forecast models as they are.

I hope this clarifies your question.

Seyma 🌷🙂

Anurag0911

@msburdak

Yes It retains the model.

You don't need to develop a new forecast model from scratch each time. Instead, you can update your data collection and retrain your existing model with the new data. Simply rerun the Forecast action to generate a new forecast based on the updated actuals.

Please note that this process produces an entirely new forecast. As a result, some users prefer utilizing the snapshotting process to gain a better understanding of the error metrics, as it generates a new forecast each time the action is executed.

I hope this helps!

msburdak

Hi Anurag,

Thanks for the answer. I am aware that the model is retained. My question is whether the model is 'retrained' every time run the forecast action. Would you be able to inform me about that?

msburdak

Thanks Anurag, that does answer it.

msburdak

Hi Anurag, just a follow up question. Is it possible to retrain an existing forecast model, without creating a new one?

seymatas1

To decide whether to train your models every month, check your monthly usage.

Seyma 🌷🙂

Anurag0911

@msburdak

As Mentioned, You have to retrain your model manually as the steps shown above. Anaplan does not support an one click solution from UX to retrain the model as that is not a requirement usually (training the model is computationally intensive process)

I hope that helps.