Skip to content

Standard modelling updates

Q2 2023

Insight PWE available in Standard Profile

From the Q2 2023 release, you can display the Insight PWE metric numerically in a standard profile. Simply use the right-click Choose your columns option to make this metric visible in your display.

Selecting the Insight PWE column

Q1 2023

Edit dimension details

The availability of the Edit dimension details option for standard modelling is useful for seeing the validation message (also available as a column), and for editing the description of an expression dimension.

Edit dimension details dialog

Note

It is not possible to modify the description of a variable dimension - this is always taken from the description given in the system definition.

Q4 2022

The concepts, charts and measures available for behavioural modelling are also available for standard modelling. Since there is no concept of "evaluation" in standard modelling, the insight definitions are all based on training data alone. This represents a change from previous release when evaluation data was used in some parts. The only chart not available in standard modelling is "Agreement PWE" since this comparesTraining and Evaluation timepoints.

Below is an example of the Insight PWE v Coverage (Dimensions) chart used for standard modelling.

The chart highlights some interesting points:

  1. Demographic variables tend to be "broad" features with high coverage.

  2. The Town variable (highlighted) has just a few insight categories, for example large towns such as London. It could be overlooked if only using Mean PWE but - as the chart below demonstrates - it stands out based on Insight PWE.

  3. The transactional variables with default bandings are appearing as niche features. This is because most people are in the same band (and hence average) and only a few people, therefore, stand out as different).

The following charts show that the default banding makes Total Booking Cost a niche feature. This is due to there being just one majority category, and only a few people stand out as being different (below average).

Custom banding offers more balanced insight, with higher 100% coverage; everybody would be given some level of positive or negative score when banded in this way.

Related topic: