SnowPro Advanced: Data Scientist (DSA-C03) Guide (2026)

The SnowPro Advanced: Data Scientist Certification validates advanced knowledge of machine learning workflows on Snowflake — feature engineering, model building, evaluation, and deployment. The exam focuses on Snowflake-native ML capabilities: Snowpark ML, DataFrame API, Feature Store, Model Registry, and ML Pipeline.

Exam Overview

Item	Details
Questions	65 questions (single choice and multiple choice)
Duration	115 minutes
Passing score	750 out of 1000
Exam fee	$375 USD
Prerequisite	Active SnowPro Core certification
Delivery	Pearson VUE (test center or online)
Certification validity	2 years
Recommended experience	2+ years of hands-on machine learning experience on Snowflake, plus Python/Scikit-learn experience

Exam Domains and Weighting

Domain	Weight	Key Topics
1. Data Preparation & Feature Engineering	25%	Snowpark DataFrame, feature transformation, Feature Store, data quality
2. Model Development	25%	Snowpark ML (model training and tuning), ML Functions, Snowpark Python UDF
3. Model Deployment & Scoring	20%	Model Registry, UDF deployment, batch inference, real-time inference
4. Model Evaluation & Monitoring	15%	Evaluation metrics, model drift detection, A/B testing, monitoring
5. ML Pipeline & Operations	15%	Pipeline automation with Tasks/DAGs, CI/CD, reproducibility

Snowpark DataFrame API

Snowpark DataFrame is an API that lets you process data with Python inside Snowflake. Because execution runs on Snowflake's compute engine, you can work with large datasets without moving data out.

Operation	DataFrame method	SQL equivalent
Read table	session.table("db.schema.table")	FROM db.schema.table
Select columns	.select(col("c1"), col("c2"))	SELECT c1, c2
Filter	.filter(col("c1") > 100)	WHERE c1 > 100
Aggregate	.group_by("c1").agg(avg("c2"))	GROUP BY c1
Join	.join(df2, "key")	JOIN df2 ON key
Write	.write.save_as_table("target")	CREATE TABLE AS SELECT

Snowpark DataFrame uses lazy evaluation. SQL execution doesn't actually happen until an action method like .collect(), .show(), or .save_as_table() is called. This lets Snowflake generate an optimal query plan after multiple transformations are chained together.

Snowpark ML

Snowpark ML is a Python library for running the entire machine learning lifecycle inside Snowflake.

Preprocessing and Feature Transformation

snowflake.ml.modeling.preprocessing：StandardScaler, MinMaxScaler, OrdinalEncoder, OneHotEncoder, LabelEncoder
Provides Scikit-learn-compatible fit/transform/fit_transform interfaces
Transformations are pushed down and executed on a Snowflake warehouse

Model Training

snowflake.ml.modeling：Scikit-learn / XGBoost / LightGBM-compatible models such as RandomForestClassifier, XGBClassifier, and LogisticRegression
Train a model by passing a Snowflake DataFrame directly to the fit() method
Training runs inside the Snowflake warehouse's Python sandbox

Hyperparameter Tuning

GridSearchCV：Evaluates every combination in a parameter grid
RandomizedSearchCV：Efficiently searches via random sampling
Cross validation also runs inside Snowflake

Feature Store

Snowflake Feature Store provides centralized management, reuse, and versioning of features.

Capability	Description
Feature View	Registers a feature definition as a SQL query or DataFrame transformation
Entity	Business entity the feature is associated with (e.g., customer_id)
Point-in-time correctness	Retrieves feature values accurate to the training timestamp (prevents data leakage)
Versioning	Tracks change history of feature definitions
Training/inference consistency	Generates both training data and inference data from the same Feature View

Model Registry

Snowflake Model Registry provides model versioning, metadata tracking, and deployment management.

Registers models as objects inside a Snowflake schema
Records metrics (accuracy, F1 score, etc.) for each version
Deploys models as UDFs to run batch inference from SQL
Integrates with Snowpark Container Services to build real-time inference endpoints

Snowflake ML Functions

Built-in machine learning functions that you can use with SQL alone, performing basic ML tasks without any Python.

Function	Use case	Input
FORECAST	Time-series forecasting	Timestamp + numeric column
ANOMALY_DETECTION	Anomaly detection	Timestamp + numeric column
CONTRIBUTION_EXPLORER	Driver analysis	Categorical + numeric columns
TOP_INSIGHTS	Data segment analysis	Categorical + numeric columns

Model Evaluation

Classification models:Accuracy, Precision, Recall, F1-score, AUC-ROC, Confusion Matrix
Regression models:MSE, RMSE, MAE, R-squared
Cross validation:Evaluate generalization performance with K-Fold Cross Validation
Model drift detection:Monitor distribution shifts in features between training and inference data

Automating the ML Pipeline

You can automate ML pipelines by combining Snowflake Tasks with DAGs (directed acyclic graphs).

Define each step (ingest → feature transformation → training → evaluation → deployment) as Tasks
Run the pipeline on a CRON schedule or via Stream triggers
Encapsulate complex ML logic in Snowpark Python stored procedures
Combine with Feature Store and Model Registry to build reproducible pipelines

Sample Question

SnowPro Advanced: Data Scientist

問題 1

Which problem does Snowflake Feature Store's Point-in-Time Correctness solve?

Slow model inference
Data leakage caused by future information leaking into training data
Excessive missing values in features
Difficulty selecting model hyperparameters

正解: B

Point-in-time correctness retrieves only the feature values that were available as of each record's timestamp when building training data. This prevents data leakage, where future information seeps into the training set, and keeps production performance from diverging from training-time performance.

Frequently Asked Questions

Does the SnowPro Advanced Data Scientist exam include Python coding questions?

You won't be asked to write and execute code, but you will read Snowpark Python DataFrame API and Snowpark ML API snippets and reason about their behavior. Expect questions like predicting the result of a chained session.table().filter().group_by() call, or selecting the right parameter settings for Snowpark ML's GridSearchCV. Basic familiarity with Pandas and Scikit-learn is also assumed.

What is the difference between Snowpark ML and Snowflake ML Functions?

Snowpark ML is a Python API library for running model training, hyperparameter tuning, and feature engineering inside Snowflake. Snowflake ML Functions, on the other hand, are built-in SQL functions (FORECAST, ANOMALY_DETECTION, CONTRIBUTION_EXPLORER, etc.) that let you perform ML tasks using only SQL. The exam includes scenarios that ask which one to use.

In what scenarios are Feature Store and Model Registry tested?

Feature Store questions cover centralized feature management, reuse, versioning, and point-in-time correctness. A typical scenario is: how do you guarantee identical feature transformations at training and inference time? Model Registry questions cover model versioning, stage management (Development/Production), and metadata tracking, with frequent scenarios like: how do you manage which model version is deployed to production?

Check what you learned with practice questions

Practice with certification-focused question sets

無料で問題を解いてみる

Author

NicheeLab Editorial Team

NicheeLab editorial team focused on data engineering and cloud certification learning. Content is structured around practical study needs and official exam domains.

SnowPro Advanced: Data Scientist Complete Guide