Metrics | Profiler

Monitor profiling metrics such as completeness, distinct counts, min/max, and type inference to detect data quality issues.

​

Profiler Metrics

Here you will find information about the supported metrics for the different types.

A Metric is a computation that Collate can run on top of a Table or Column to receive a value back. They are the primary building block of Collate’s Profiler.

• Metrics define the queries and computations generically. They do not aim at specific columns or database dialects. Instead, they are expressions built with SQLAlchemy that should run everywhere.
• A Profiler is the binding between a set of metrics and the external world. The Profiler contains the Table and Session information and is in charge of executing the metrics.

On this page, you will learn all the metrics that Collate currently supports and their meaning. The namings are based on the definitions on the JSON Schemas.

Tip: For the columnProfile definition, see the table schema. The metrics are implemented in the profiler metrics source.

The namings are based on the definitions on the JSON Schemas.

​

Table Metrics

Those are the metrics computed at the Table level.

​

Row Count

It computes the number of rows in the Table.

​

Column Count

Returns the number of columns in the Table.

​

Column Names

Returns the list of column names in the Table.

​

System Metrics

System metrics provide information related to DML operations performed on the table. These metrics present a concise view of your data freshness. In a typical data processing flow tables are updated at a certain frequency. Table freshness will be monitored by confirming a set of operations has been performed against the table. To increase trust in your data assets, Collate will monitor the INSERT, UPDATE and DELETE operations performed against your table to showcase 2 metrics related to freshness (see below for more details). With this information, you are able to see when a specific operation was last perform and how many rows it affected.

These metrics are available for BigQuery, Redshift and Snowflake. Other database engines are currently not supported so the computation of the system metrics will be skipped.

​

DML Operations

This metric shows all the DML operations performed (INSERT, UPDATE, DELETE) against the table in a timeseries fashion.

​

Rows Affected by the DML Operation

This metric shows the number of rows that were affected by a DML operation (INSERT, UPDATE, DELETE) over time.

​

Column Metrics

Metrics that Collate runs for all columns.

Note that Collate does not currently support complex types such as ARRAY or STRUCT. The implementation will come down the road.

​

Values Count

It is the total count of the values in the column. Ignores nulls.

​

Values Percentage

Percentage of values in this column vs. the Row Count.

​

Duplicate Count

Informs the number of rows that have duplicated values in a column. Computed as count(col) - count(distinct(col)).

​

Null Count

The number of null values in a column.

​

Null Proportion

It shows the ratio of null values vs. the total number of values in a column.

​

Null Missing Count

The number of null values plus empty-string values in a column that can contain strings. Unlike Null Count, this also treats empty strings as missing values. The exact empty-string behavior depends on the profiler backend and column type.

Null Missing Count is not part of the default profiler metric set. In the Profiler Settings metrics picker, selecting it currently has no effect because the backend registers it as nullMissingCount but reports its result as nullCount. Use a custom profiler configuration only if you can account for that result name until the identifier is corrected.

​

Unique Count

The number of unique values in a column, those that appear only once. For example, [1, 2, 2, 3, 3, 4] => [1, 4] => count = 2.

​

Unique Proportion

Unique Count / Values Count.

​

Distinct Count

The number of different items in a column. For example, [1, 2, 2, 3, 3, 4] => [1, 2, 3, 4] => count = 4.

​

Distinct Proportion

Distinct Count / Values Count.

Count in Set, Like Count, Case-Insensitive Like Count, Not Like Count, Regex Count, and Not Regex Count each require an additional runtime property that isn’t exposed in the Profiler Settings metrics picker. When listed without values or expression, the ingestion profiler skips them. Configure them in a custom profiler workflow with add_props(values=...) for Count in Set or add_props(expression=...) for the pattern-based metrics. In the Collate Spark implementation, Case-Insensitive Like Count currently uses case-sensitive regular-expression (rlike) matching, so do not rely on case-insensitive matching until that implementation is corrected.

​

Count in Set

Counts the number of values in a column that match a specified set of values. Requires the set of values to check against.

​

Like Count

Counts the number of values in a column matching a specified SQL LIKE pattern. Requires a pattern.

​

Case-Insensitive Like Count

Counts the number of values in a column matching a specified pattern, ignoring case. Requires a pattern. The Collate Spark implementation currently evaluates the pattern with case-sensitive regular-expression (rlike) semantics; see the warning above.

​

Not Like Count

Counts rows whose value does not match a specified SQL LIKE pattern. SQL-backed runs count NULL values as non-matches, while Spark-backed runs do not count NULL results from the negated predicate.

​

Regex Count

Counts the number of values in a column matching a specified regular expression.

​

Not Regex Count

Counts rows whose value does not match a specified regular expression. SQL-backed runs count NULL values as non-matches, while Spark-backed runs do not count NULL results from the negated predicate. The current Pandas implementation counts matching values instead, so do not use this metric with a Pandas profiler until that implementation is corrected.

​

Min

Only for numerical values. Returns the minimum.

​

Max

Only for numerical values. Returns the maximum.

​

Min Length

Only for concatenable values. Returns the minimum length of the values in a column.

​

Max Length

Only for concatenable values. Returns the maximum length of the values in a column.

​

Mean

• Numerical values: returns the average of the values.
• Concatenable values: returns the average length of the values.

​

Median

Only for numerical values. This is currently not supported in MySQL.

​

Sum

Only for numerical values. Returns the sum of all values in a column.

​

Standard Deviation

Only for numerical values. Returns the standard deviation.

​

Histogram

The histogram returns a dictionary of the different bins and the number of values found for that bin. It will be computed only if the Inter Quartile Range value is available.

​

First Quartile

Only for numerical values. Middle number between the smallest value and the median.

​

Third Quartile

Only for numerical values. Middle number between the median and the greatest value.

​

Inter Quartile Range

Only for numerical values. Difference between the third quartile and the first quartile.

​

Nonparametric Skew

Measure of skewness of the column distribution. Nonparametric skew is computed as follows.

$$S = \frac{\mu-\tilde{\mu}}{\sigma}$$

Where

$$\mu = mean\\ \tilde{\mu} = median\\ \sigma = standard deviation\\$$

​

Grant Access to User for System Metrics

Collate uses system tables to compute system metrics. Find the required access as well as more details for your database engine below.

​

Snowflake

Collate uses the QUERY_HISTORY_BY_WAREHOUSE view of the INFORMATION_SCHEMA to collect metrics about DML operations. The QUERY ID is passed to the RESULT_SCAN function to get the number of rows affected by the operation. Ensure the user running the profiler workflow has access to this view and this function.

Collate will look at the past 24 hours to fetch the operations that were performed against a table.

Important For the Snowflake system, the system will parse the DDL query and attempt to match database, schema, and table name to entities in Collate. If the DDL query does not include all 3 elements Collate will not be able to ingest this metric.

​

Redshift

Collate uses the following to fetch DML operations as well as the number of rows affected by these operations.

• SVV_TABLE_INFO, STL_INSERT, and STL_DELETE in Provisioned Cluster
• SYS_QUERY_DETAIL in Serverless instance

Ensure the user running the profiler workflow has access to these views and tables.

Collate will look at the previous day to fetch the operations that were performed against a table.

​

BigQuery

BigQuery uses the JOBS table of the INFORMATION_SCHEMA to fetch DML operations as well as the number of rows affected by these operations. Make sure your data location is properly set when creating your BigQuery service connection in Collate.

Collate will look at the previous day to fetch the operations that were performed against a table filter on the creation_time partition field to limit the size of data scanned.

​

Reach out!

Is there any metric you’d like to see? Open an issue or reach out on Slack.