Skip to end of metadata
Go to start of metadata

You are viewing an old version of this page. View the current version.

Compare with Current View Page History

« Previous Version 2 Next »

This is what is new within the Optimization module, specifically the model and model class. We've done some cleaning and removed a few things, so you'll notice that the Optimization module is now cleaner and has many updates. You'll still see the main features, including the models that will be listed here.

Good to know: The Optimization Engine page has been removed as the engine has become a direct part of other models. You can still enable it until completely removed after the next release (12.0)

Good to know: Pricing Guidance is no longer used with all model types, and it will be deprecated. It's still possible to use it, and existing projects using it are still valid. However, we highly recommend not starting any new project using Pricing Guidance for all model types.

Good to know: The configuration of model classes is no longer part of the menu; it's now part of the administrative and configuration module.

Multifactor elasticity

The multifactor elasticity model forecasts the quantity that would be sold based on past transactions in the data. Moreover, some variations of prices to understand their impact on quantity have been made.

LEARN MORE: Understand models and how you can configure them, here.

Pricefx built an elasticity model related to all the information you can get, all the features and fields in your data set, to understand the relationship between price and quantity for your specific case. This is used for Shelf Price positioning and to understand the dynamics of the market.

LEARN MORE: To know more about how Shelf Price works, click here. 

In this multifactor elasticity, the possibility to define the number of periods to forecast (up to 15 periods, meaning we can forecast up to 15 weeks) was introduced in the Paper Plane release as it is used for Markdown, another optimization accelerator. Markdown is perfect when moving excess products in stock, when products of a specific kind need to be merged and to reduce stock. This model can also be used to position the best price to reduce the stock.

LEARN MORE: Find out what Markdown can do for you, here.

Additional parameters

In the Paper Plane 11.0 release some additional parameters were added. The range of the prediction for elasticity can be defined. For example, one can consider a 20% change in price good enough for Shelf Price, but for Markdown, where the price can be aggressively decreased, it should probably go up to 50% or even 60%. 

By clicking the Model Parameter Configuration, you can tune the values and define the number of trials you would like to do, as well as the number of cross-validation and splits. This means taking some data out, computing it, checking with the data that's been taken out, and seeing if that model makes sense or not. It's something you typically do at least once when you start working on it, and you probably have to update it every three to six months depending on your markets.

Good to know: That's not how we define the forecast, but it helps to define the best parameters for it. It can be time-consuming and take anywhere from 10 to 20 hours, depending on the size of the model. So be prepared to be patient in that case.

Model training

In the next step, Model Training, there are several new refinements. The chart on Feature Importance that has not been there before, is showing the main drivers to define the forecast with all the aspects you may want to consider. There is also a new table with additional metrics to understand the accuracy of the model.

In the model predictions, you will see that there is a demand forecast for all, and for specific products. 

The main target of this model is to make a price change. So for the last week price, we make a price change for instance by 20% more and 20% less, and then we see what the model will predict. We fit on top of this a curve to understand what is the relationship between price and quantity.

LEARN MORE: To learn how you can train the model, click here.

 

  • No labels