The latest release brought many great features to Einstein Discovery. We are going to summarise some of them in this blog article.
We now have model cards that can be used to help model onboarding and documentation. A model card shows data statistics and explanations for usage, design, limitations, and anything else that we deem important. The cards are created in Model Manager in an easy-to-understand point and click interface. The model cards can be downloaded in PDF format and presented on a page in Lightning App Builder.
Missing values in data can now be imputed using statistical imputation. This procedure allows for automatic replacement of missing values based on values derived from another subset of the data. Observations with missing values can then be used during analysis. This feature can be found and enabled in story settings.
The next feature is an addition to story settings which is a fuzzy matching variable transformation for categorisation. This avoids treating variations in values as distinct categories if they have a common root category. We can use this to smooth differences in abbreviations, plurals, capitalisation, and spellings. Colour, color, colours, and colour can all be treated as the same. This transformation is applied at prediction time, meaning we do not need to cleanse the data source during data prep to gain better insight with the fuzzy matching variable transformation.
Finally in our list, we can now use Insights API to embed insights from our story into any website, application, or dashboard. The Einstein Discovery system permission unlocks the ability to share the quantitative insights and their narrative descriptions with broader audiences and enables informed decision making.
This list of features is in no way exhaustive as Salesforce Einstein Discovery is continuously powering up. If you need to learn more about optimal AI and BI strategies you will find more information on our website where you can also contact us for assistance.