1.2.0

Dec 13, 2023

Results of the Model Predictions step are now displayed in two tabs: Overview and Details. There are additional metrics, such as histograms showing the distribution of R² scores, simple elasticities and the k parameter of the sigmoid curve fit. A table with elasticity data for each product, store, and date of the forecast is also included.

PFPCS-7137

In the Definition step, there is a dedicated (optional) mapping for list price which is then used to calculate discount %. This allows for special handling of discounts, such as converting absolute discounts to percentages, recalculating discounts when generating the elasticity curves etc.

PFPCS-7360

It is possible now to map past and future events that should be part of the model features and forecast. In Additional Sources step, you can add general calendar events (defined by date key) and specific promotions / local events (defined by date and secondary/tertiary keys). You can also add data to be used when making the forecast (e.g. future prices, discounts).

PFPCS-7363

In the Model Training step on the Elasticity Settings Tab, it is possible now to set fallback elasticity for products with low transaction counts or low quality elasticity curves. The fallback provides a simple elasticity for these products based on the elasticities of other products at the defined fallback level. 

PFPCS-7441

The elasticity calculation can be disabled to save on computation time. To turn it on/off, go to Model Training step > Elasticity Settings tab and use the "Calculate elasticities" checkbox. If it is disabled, only forecast is calculated in the final model step.

PFPCS-7560

In the Model Training step on the Elasticity Settings Tab, there is a new option "Adjust elasticity curve fit based on last known price prediction" which forces the calculated elasticity curve to go through the model prediction at the last known price. This helps increase the reliability of elasticities for products/datasets with low price variation and reduce bias in outputs.

PFPCS-7599

Default preferences for layout are now part of the accelerator release.

PFPCS-7604

Dimensions Product and Date were added to the forecast and elasticity_factors tables, so that the data can be analyzed elsewhere on the platform.

PFPCS-7653

When installing the accelerator from PlatformManager, engines configuration steps are now the first steps of the accelerator installation. This is to prevent errors when trying to run a model while the accelerator installation has not been completed.

PFPCS-7702

Some seasonality features are missing in daily models in the forecast generated in the Model Predictions step.

PFPCS-7201

When using the automatic model tuning, the combination of hyperparameters in a particular trial can fit the data so poorly that it produces predictions of infinite values. When this happens the scoring function fails with the error message "Input contains infinity or a value too large for dtype('float64')." and prevents further trials.

PFPCS-7257