Skip to main content

Best Practices

What are some tips for getting started?​

We recommend that the testing data model have at least one time hierarchy and one standard hierarchy. This will better illustrate the breadth of AI-Link’s capabilities for programmatically interacting with AtScale data models.

The following command installs the latest version of AI-Link:

pip install atscale --upgrade

See our warehouse extras for more on maintaining packages via pip that are specific to your data warehouse.

What if I have a really large dataset? Is there a way to sample data?​

Many of our customers manage very large datasets in their warehouses. While AtScale does help optimize query performance, there is a limit to the number of rows you can fetch per query (which is configurable in your AtScale settings) and a limit to the amount of data your local kernel can handle.

To address this, we recommend sampling your data with a get_data call using the function’s filtering parameters. One way of doing this is by filtering by a categorical variable. For instance, this call returns average sales data at the month level, but only for stores in Missouri:

my_data_model.get_data(
    feature_list=[“month”, “average_sales”],
    filter_in={
        “state”: [“Missouri”],
    },
)

and this call also gets month-wise average sales data, but only after January 23rd, 2024:

my_data_model.get_data(
	feature_list=[“date”, “month”, “average_sales”],
	filter_greater={
		“date”: “01-23-24”,
    },
)