ML-Powered Functions

These analysis tools give you automated predictions and insights into your data using machine learning.

What Are ML-Powered Functions?

Snowflake’s ML-powered functions use machine learning to detect patterns in your data. We provide an appropriate type of model for each feature, so you don’t have to be a machine learning developer to take advantage of them. All you need is your data.

Time-Series Functions

These features train a machine learning model on your time-series data to determine how a specified metric (for example, sales) varies over time and relative to other features of your data. The model then provides insights and predictions based on the trends detected in the data.

  • Forecasting predicts future metric values from past trends in time-series data.

  • Anomaly Detection flags metric values that differ from typical expectations.

  • Contribution Explorer helps you find dimensions and values that affect the metric in surprising ways.

Cost Considerations

When you use ML-powered functions, you incur storage and compute costs. These costs vary depending on the feature used and the quantity of data used in training and prediction.

The storage costs you incur reflect storage of the ML model instances created during the training step. To view the objects associated with your model instance, navigate to your account usage views (e.g., ACCOUNT_USAGE.TABLES and ACCOUNT_USAGE.STAGES). These objects will appear with null database and schema columns. The instance_id column, however, will be populated and indicates that these objects are contained inside a model instance. These objects are fully managed by the model instance, and you cannot access or delete them separately. To reduce storage costs associated with your models, delete unused or obsolete models.

See Understanding Compute Cost for general information on Snowflake compute costs.