Extracts can be created in multiple ways, as follows:
- Select Extract on the Data Source screen as follows. The Edit... link will allow you to configure the extract:
- Select the data source from the Data menu, or right-click the data source on the data pane and select Extract data.... You will be given a chance to set configuration options for the extract, as demonstrated in the following screenshot:
- Developers may create an extract using the Tableau Data Extract API. This API allows you to use Python or C/C++ to programmatically create an extract file. The details of this approach are beyond the scope of this book, but documentation is readily available on Tableau's website.
- Certain tools, such as Alteryx or Tableau Prep, are able to output Tableau extracts.
When you first create or subsequently configure an extract, you will be prompted to select certain options, as shown here:
You have a great deal of control when configuring an extract. Here are the various options, and the impact your choices will make on performance and flexibility:
- You may optionally add Extract filters, which limit the extract to a subset of the original source. In this example only, records where Region is Central or South and where Category is Office Machines will be included in the extract.
- You may aggregate an extract by checking the box. This means that data will be rolled up to the level of visible dimensions and, optionally, to a specified date level, such as year or month.
Visible fields are those that are shown in the data pane. You may hide a field from the Data Source screen or from the data pane by right-clicking a field and selecting Hide. This option will be disabled if the field is used in any view in the workbook. Hidden fields are not available to be used in a view. Hidden fields are not included in an extract as long as they are hidden prior to creating or optimizing the extract.
- In the preceding example, if only the Region and Category dimensions were visible, the resulting extract would only contain two rows of data (one row for Central and another for South). Additionally, any measures would be aggregated at the Region/Category level and would be done with respect to the Extract filters. For example, Sales would be rolled up to the sum of sales in Central/Office Machines and South/Office Machines. All measures are aggregated according to their default aggregation.
- You may adjust the Number of Rows in the extract by including all rows or a sampling of the top N rows in the dataset. If you select all rows, you can indicate an incremental refresh. If your source data incrementally adds records, and you have a field such as an identity column or date field that can be used reliably to identify new records as they are added, then an incremental extract can allow you to add those records to the extract without recreating the entire extract. In the preceding example, any new rows where Row ID is higher than the highest value of the previous extract refresh would be included in the next incremental refresh.
Incremental refreshes can be a great way to deal with large volumes of data that grow over time. However, use incremental refreshes with care, because the incremental refresh will only add new rows of data based on the field you specify. You won't get changes to existing rows, nor will rows be removed if they were deleted at the source. You will also miss any new rows if the value for the incremental field is less than the maximum value in the existing extract.