Source dataset

A source dataset is a logical table from the data warehouse. When you connect to the data warehouse, you can map a database table, database view, or SQL query from the data warehouse to a source dataset within Analytics without physically transferring or moving data to Analytics. You can then use the source dataset to execute analysis or create derived datasets.

The following are the different sections available for dataset definition:

Analytics displays a default SQL query on the definition page based on your data selections, including the connection source and contents. You can modify the suggested SQL command to shape the data as needed. Click Run Query and preview the data to ensure the SQL statement returns the expected information.

This section lets you inspect the properties of individual columns. This section defines the following properties:

A primary key is a column or set of columns in a dataset that uniquely identifies each row within the dataset. While configuring a primary key is optional in Optimizely Analytics, it is a highly recommended best practice for ensuring the quality and reliability of your data.

The primary key's main job is to ensure that every row is unique. Without it, you could have duplicate rows that inflate your metrics. For example, if you have two identical rows for the same user, you might incorrectly count them as two separate users, leading to flawed analysis and business decisions. A primary key acts as a safeguard against this.

To define a primary key, ensure that a dataset within Analytics does not include fully duplicated rows, because you cannot establish a primary key for such a dataset.

Health check

The primary key health check is a tool designed to verify that the primary key is functioning correctly, letting you proactively detect errors and build confidence in your data foundation.

Use Check health in the Primary Keys tab to evaluate a given set of columns designated as a primary key to determine how many distinct key values have duplicate rows.

For datasets identified as event streams, the check automatically filters data to the last 60 days. To ensure predictable performance and cost efficiency, the scan is limited to a maximum of 10 million rows.

There are three different health statuses possible, and they are as follows:

In this section, you can define new derived columns and cohorts of rows that represent logical groupings within the respective tabs. This is an optional step. Learn more about cohorts and derived columns.

The Related datasets tab lets you define logical relationships between datasets, similar to how tables relate in a data warehouse. You can combine event and attribute data for richer analysis without understanding the underlying data structure. Analytics uses these relationships to determine which datasets to join in an exploration and only shows relevant options in the UI. For example, using common user ID columns, you can form a relationship between an events dataset and a users dataset.

The Check health option in the Related Datasets tab evaluates a defined relationship between two datasets. It identifies how many distinct values on the many side of the relationship do not have a corresponding match on the one side. This helps uncover broken links or orphaned records that could impact data integrity.

For datasets identified as event streams, the check automatically filters data to the last 60 days. To ensure predictable performance and cost efficiency, the scan is limited to a maximum of 10 million rows.

There are three different health statuses possible, and they are as follows:

Sampling keys are exclusive to an actor dataset and let you use sampled data when creating an exploration using this source dataset.

To configure sampling,

Semantics let you annotate your dataset as an actor, decision stream, event stream, or SCD.

Represents entities that perform actions. When Actor is chosen, click Add Preferred Property and add properties.

Captures decision events such as experiment bucketing. If Decision Stream is chosen, you must configure the following fields:

Contains time-series actions performed by users or other actors. This is often behavioral data, such as clicks, page views, or transactions. When you select Event stream, click Add Event Stream and configure event and actor properties.

Custom partition time column improves query performance by applying exact filters on event time and relaxed filters on a mapped partition column, based on a set time skew.

Use the Index Time Range field to apply a time filter when indexing distinct column values and JSON fields, overriding the app-level setting if specified. You can reset this to the default value when required.

The Session Identifier field in the Event Stream Properties section determines how events are grouped for each actor. It is crucial for defining what constitutes a session for a user within your data. The chosen session identifier, such as a user_id or session_id, should be unique and consistent across your tables to ensure accurate event grouping and analysis.

You can see the events in the dataset when you click See all events.

SCD lets you handle historical changes in your data more effectively, with minimal configuration. SCD2 datasets require only a many-to-one relationship with the actor dataset. When you configure this, Optimizely Analytics auto-joins any event stream that joins with the actor dataset (for example, Users) with the SCD2 table; you do not have to perform any additional setup.

Model this type with the following required fields:

You can view and access your dataset on the Datasets page when you save it.

Any event the user triggers becomes an active event. For example, a user click is an active event. Analytics lets you define active events by identifying a cohort on any event stream. To define an active event, complete the following:

This cohort displays in the active events drop-down list in the Semantics tab of your dataset.

You can use active events inside features like funnels when you define active events. While defining the stages in a funnel, Analytics lets you set stages in which only the active events display. In Funnels, you can only see active events if your event stream has an annotation for active events.