Learning Objectives
By the end of this chapter, you will be able to:
- Define machine learning concepts for data science workloads.
- Outline machine learning problem types.
- Summarize the machine learning lifecycle.
- Define statistical concepts for data science.
Executive Summary
- Supervised learning and unsupervised learning differ by whether a target column exists in the training data, and that single attribute drives both algorithm selection and the Snowflake feature you reach for (Cortex
CLASSIFICATIONandFORECASTfor supervised,ANOMALY_DETECTIONand clustering through Snowpark ML for unsupervised). - Problem type selection follows from the target's data type (continuous, binary, categorical, temporal, image) and dictates the loss function, the evaluation metric, and whether the workload fits a virtual warehouse or requires the Container Runtime with GPU instances.
- The lifecycle is sequential in description but iterative in practice: data collection, exploration, feature engineering, training, registry-based deployment, and monitoring all happen inside the Snowflake account when teams use Notebooks, Feature Store, Model Registry, and Snowflake observability tables together.
- Statistical literacy is not optional at the advanced tier: distribution shape decides imputation strategy, the central limit theorem justifies confidence intervals on aggregate metrics, and the choice between Z and T tests depends on sample size and variance knowledge, both of which the exam tests in scenario form.
Assumptions
- The reader holds SnowPro Core and has built at least one end-to-end model on Snowflake using either Snowpark Python or Cortex ML Functions.
- All SQL examples assume Snowflake's standard dialect and a current edition of Snowpark Python; preview-only features are labeled inline.
- Naming convention throughout the chapter: database
RETAIL_DSC_DB, schemasRAW,FEATURES,ML_DEV,ML_PROD, warehousesDSC_TRAIN_WHandDSC_SERVE_WH, primary tablesCUSTOMERS,ORDERS,SESSIONS,CHURN_LABELS. - Region availability of Cortex ML Functions varies; consult the Snowflake region matrix before relying on a specific function in production design.