In the world of predictions we try to summarize how well we do with some over-arching metrics such as P-values
, or precision and recall
While a useful shorthand, these metrics are abstract and don't provide all relevant details (how could they). So as an alternative, it's nice to be able to pull-out a time-series database (Influxdb) and visualization platform (Grafana) and plot out our predictions followed by the actual results. Sometimes a simple graph can be worth a thousand KPIs.
The above is a trivial example where we pull the weather forecast from a public API, and later the actual weather from sensors managed by our friends at Alvasys
. The red and blue values are predictions, and the green values are actuals.
Now imagine that instead of weather data, you can visualize your ML/AI predictions alongside the actual results. Influxdb and Grafana make it easy to query by time, model type or any dynamic metric you add as metadata. These graphs can be a precious complement to your over-arching metrics and help you form a more holistic understanding of your model performance.