02 navigating the ui
The Tecton Web UI allows you to review the current contents of your production Feature Store. For example, the UI allows you to view the features you and your colleagues have placed in the Feature Registry, understand the freshness and frequency of those features as they are updated, and see how features are being consumed by model services and applications in production.
Exploring the Feature Registry
The Feature Registry exposes all of the relevant information around the features your organization has developed and may be using in production. This is where data scientists can share features: reviewing the Feature Registry gives data scientists the transparency they need to understand whether a feature should be included in their own model. It also helps data scientists understand the health and performance of their own features.
Exploring the different FeaturePackage tabs
In the Web UI, click on the Features tab on the left hand side, and then the
partner_ctr_performance FeaturePackage. From there, you can review the different tabs:
- Overview: This tab contains the owner and a code example of how we can fetch feature values inside of your notebook. At the bottom, you can also review the features that are inside of this feature package.
- Transformations: This tab highlights that the feature is built off of the ad impressions batch data source. It also shows the transformations associated with this feature, including the specific SQL logic used to generate the output.
- Materialization: This tab contains information about the actual materialization of this feature. It shows the historical lineage of the feature values (start time), how frequently values are materialized, the health of the features, and the flow of the overall data.
- Feature Services: This tab contains information about the which production models are consuming the feature's data as an input for prediction.
- Monitoring: This tab contains information about the operational performance for feature generation - including ingestion rate, latencies for serving online, and the writes per second into production.
- Statistics: This tab contains summary statistics for all of the features generated in this feature package - that includes the prevalence of nulls, as well as mean, min, max, and standard deviations.
In Tecton, FeatureServices are used to group together features that create a model object. By defining a FeatureService, a data scientist can use that object to generate training sets in one-line method calls, and fetch those features' values as of "now" for inference via an API call.
In the Web UI, you can use the Services tab to understand which features are included, refer to the serving endpoint provided, and review monitoring statistics to see how the endpoint is being used and whether you are meeting the SLOs you have for your model and customers.
Exploring the different FeatureService tabs
In the Web UI, click on the Services tab on the left hand side, and then the
ctr_prediction_service FeatureService. From there, you can review the different tabs:
- Overview: This tab contains the owner and an example
curlfor calling the production feature values. At the bottom, you can also review the features that are inside of this FeatureService.
- Materialization: This tab contains information about the materialization health of the features being used in the FeatureService.
- Pipeline: This tab shows the lineage of all features in the FeatureService, from the source data to how it is stored and where it is being consumed.
- Monitoring: This tab contains information about the operational performance of the serving endpoint - for example, the distribution of latencies and the throughput (QPS) over time.
- Statistics: This tab contains summary statistics for all of the features used by this FeatureService, including the prevalence of nulls, as well as mean, min, max, and standard deviations.