Custom Metric Syntax
Breakdown of the Arize Query Language
Custom Metric Syntax Overview
Custom metrics use an Arize SQL-like query language.
SELECT
SELECT
Every query starts with a SELECT
statement.
FROM
FROM
The FROM
statement should beFROM model
, where model
refers to the model that the custom metric was created under.
Constants
A constant can be a number or string. Strings use single quotes like this: 'this is a string'
. Numbers can be floats or integers.
Expressions
An expression is a statement composed of simple math or boolean operators. An expression does not have an aggregation function within it. Expressions can include nested expressions, for example, A * (B + C). Expressions return the same number of rows as those that go into it.
Dimensions
A dimension is any feature, tag, prediction, actual, or span property of a model. A dimension can either be of type string or of type numeric.
Feature
feature_name
or "feature name"
Features can be referenced directly or use double quotes.
Note: if there are spaces in your feature name, you must use double quotes.
Tag
tag_name
or "tag name"
Tags can be referenced directly or use double quotes.
Note: if there are spaces in your tag name, you must use double quotes.
Prediction
categoricalPredictionLabel scorePredictionLabel numericPredictionLabel
For scored categorical models, there is categoricalPredictionLabel
and scorePredictionLabel
For numeric models, there is only numericPredictionLabel
Actual
categoricalActualLabel scoreActualLabel numericActualLabel
For scored categorical models, there is categoricalActualLabel
and scoreActualLabel
For numeric models, there is only numericActualLabel
Span Property
span_property
or "span property"
or "span. property"
Span properties can be referenced directly or use double quotes.
Note: if there are spaces or periods in your span property name, you must use double quotes.
Comments
Comments can be added to your query in one of two ways:
Single line comment (shortcut is CMD + /
)
Multi-line comments
Last updated