How to create a new Expectation Suite using the CLI

Show Docs for V2 (Batch Kwargs) API

While you could hand-author an Expectation Suite by writing a JSON file, just like with other features it is easier to let the CLI save you time and typos. Run this command in the root directory of your project (where the init command created the great_expectations subdirectory):

great_expectations suite new

This command prompts you to name your new Expectation Suite and to select a sample Batch of data that the suite will eventually describe. An empty suite is created and added to your project. Then it creates a jupyter notebook for you to start creating your new suite. The command concludes by opening the newly generated jupyter notebook.

If you wish to skip the automated opening of jupyter notebook, add the –no-jupyter flag:

great_expectations suite new --no-jupyter

Show Docs for V3 (Batch Request) API

While you could hand-author an Expectation Suite by writing a JSON file, just like with other features it is easier to let the CLI save you time and typos.

Simply run this command and follow the prompts:

great_expectations --v3-api suite new

Alternatively, if you wish to skip some steps by providing flags please follow the rest of this document. We will walk through several available options for how to do this.

1. The --interactive mode denotes the fact that you are interacting with your data. In other words, you have access to a data source and can specify a Batch of data to be used to create Expectations against. --manual mode still allows you to create expectations (e.g., if you already know enough about your data, such as the various columns in a database table), but you will not be able to run validations, until you specify a Batch of data, which can be done at a later point; in fact, you can switch back and forth between the interactive and manual modes, and all your expectations will be intact.

Run this command in the root directory of your project (where the init command created the great_expectations subdirectory:

great_expectations --v3-api suite new --interactive

This command prompts you to select a datasource, a data connector, and a data asset so as to identify a sample Batch of data the suite will eventually describe. If there are unique choices (e.g., only one data connector in your datasource configuration), then Great Expectations will automatically select it for you (to speed up the process).

Finally, unless you specify the name of the Expectation Suite on the command line (using the --expectation-suite option), the command will ask you to name your new Expectation Suite and offer you a default name to simply accept, or provide your own.

Then an empty suite is created and added to your project.

Then Great Expectations creates a jupyter notebook for you to start creating your new suite. The command concludes by opening the newly generated jupyter notebook.

  1. If you wish to skip the automated opening of jupyter notebook, add the --no-jupyter flag:

great_expectations --v3-api suite new --interactive --no-jupyter

or in the manual mode:

great_expectations --v3-api suite new --manual --no-jupyter
  1. One of the easiest ways to get starting in the interactive mode is to take advantage of the –profile flag:

great_expectations --v3-api suite new --interactive --profile

This instructs Great Expectations to inspect your Batch of data and suggest the initial set of Expectations. When the notebook opens (or when you open it later if the above command is used with the --no-jupyter flag), you can edit these expectations. For example, you might tighten some tolerances, compared to the initial values, based on your knowledge of your dataset. After that, you can review and save your Expectation Suite and run validations using it.

Here are some general guidelines for taking advantage of the profiling capability for bootstrapping your Expectation Suite.

Within the notebook, run the first cell in the notebook that loads the data. (This step was explained above.)

The next code cell in the notebook presents you with a list of all the columns found in your selected data:

ignored_columns = [
    "Name",
    "Age",
    "Address",
    "Occupation",
    ...
]

By default, all columns are ignored. To select which columns you want to be profiled for generating Expectations on, simply comment them out to include them.

The next code cell is where you will configure and instantiate your profiler, and build your suite. You can leave these defaults as-is for now - learn more about the available parameters here.

When you run this cell and build your suite, you will see a list of the expectations included by column. At this point, you may also make modifications to the ignored_columns or the profiler, and re-run the cell.

profiler = UserConfigurableProfiler(
    profile_dataset=validator,
    excluded_expectations=None,
    ignored_columns=ignored_columns,
    not_null_only=False,
    primary_or_compound_key=False,
    semantic_types_dict=None,
    table_expectations_only=False,
    value_set_threshold="MANY",
)
suite = profiler.build_suite()

Finally, run the next few code cells to see the automatically generated Expectation Suite in Data Docs.

Because the profiler-based expectations are too permissive (i.e., lax tolerances), you will want to edit this suite to tune the parameters and make any adjustments such as removing Expectations that don’t make sense for your use case. You can iterate on included and excluded columns and Expectations to get closer to the Expectation Suite you want.

Important

The Suites generated by the profiler are not meant to be production suites – they are the initial estimates to build upon.

Great Expectations will choose which expected values for Expectations might make sense for a column based on the type and cardinality of the data in each selected column.

You will definitely want to edit the Suite to fine-tune it after auto-generating it with the --profile flag.

When in the interactive mode, the initialization cell of your jupyter notebook will contain the batch_request dictionary. You can convert it to JSON and save in a file for future use. The contents of this file would look like this:

{
    "datasource_name": my_datasource",
    "data_connector_name": "my_data_connector",
    "data_asset_name": "my_asset"
}

You can then utilize this saved batch_request (containing any refinements you may have made to it in your notebook) and skip the steps of selecting its components:

great_expectations --v3-api suite new --interactive --batch-request my_saved_batch_request_file.json

Unless you specify the name of the Expectation Suite on the command line (using the --expectation-suite MY_SUITE syntax), the command will ask you to name your new Expectation Suite and offer you a default name for you to simply accept, or provide your own.

You can extend the previous example to specify the name of the Expectation Suite on the command line as follows:

great_expectations --v3-api suite new --expectation-suite my_suite --interactive --batch-request my_saved_batch_request.json

To check the syntax, you can always run the following command in the root directory of your project (where the init command created the great_expectations subdirectory:

great_expectations --v3-api suite new --help