In this Snowpark 101 for Data Science Quickstart guide, you will be help the fictitious food truck company, Tasty Bytes, to increase sales by training a model to provide location recommendations to truck drivers. You will use Snowpark for Python to prepare data, train a model, and deploy the model. Once deployed, you will create an application prototype using Streamlit to demonstrate how truck drivers could use the model to find the best location to park for an upcoming shift.

What is Snowpark?

Snowpark allows developers to query and create data applications in Python, Java, and Scala through APIs and DataFrame-style programming constructs that run on Snowflake's elastic engine. Learn more about Snowpark.

What is Streamlit?

Streamlit (acquired by Snowflake in March 2022) is a Python library that makes it easy to create and share custom web apps. Learn more about Streamlit.

What You Will Learn


What You Will Build


You will use Snowsight, the Snowflake web interface, to:

Tasty Bytes operates food trucks in cities across the globe with each truck having the ability to choose two different selling locations per day. The locations are mapped to SafeGraph points-of-interest. You want to join latitude and longitude from the SafeGraph Marketplace data to your shift sales data to use as features in model training.

Step 1 - Acquiring SafeGraph POI Data from the Snowflake Marketplace

Step 2 - Creating Objects, Loading Data, and Joining Data

USE ROLE accountadmin;

-- create a development database for data science work
CREATE OR REPLACE DATABASE frostbyte_tasty_bytes_dev;

-- create raw, harmonized, and analytics schemas
-- raw zone for data ingestion
CREATE OR REPLACE SCHEMA frostbyte_tasty_bytes_dev.raw;
-- harmonized zone for data processing
CREATE OR REPLACE SCHEMA frostbyte_tasty_bytes_dev.harmonized;
-- analytics zone for development

-- create csv file format
CREATE OR REPLACE FILE FORMAT frostbyte_tasty_bytes_dev.raw.csv_ff 
type = 'csv';

-- create an external stage pointing to S3
CREATE OR REPLACE STAGE frostbyte_tasty_bytes_dev.raw.s3load
COMMENT = 'Quickstarts S3 Stage Connection'
url = 's3://sfquickstarts/frostbyte_tastybytes/'
file_format = frostbyte_tasty_bytes_dev.raw.csv_ff;

-- define shift sales table
CREATE OR REPLACE TABLE frostbyte_tasty_bytes_dev.raw.shift_sales(
	location_id NUMBER(19,0),
	city VARCHAR(16777216),
	date DATE,
	shift_sales FLOAT,
	shift VARCHAR(2),
	month NUMBER(2,0),
	day_of_week NUMBER(2,0),
	city_population NUMBER(38,0)

-- create and use a compute warehouse
USE WAREHOUSE tasty_dsci_wh;

-- ingest from S3 into the shift sales table
COPY INTO frostbyte_tasty_bytes_dev.raw.shift_sales
FROM @frostbyte_tasty_bytes_dev.raw.s3load/analytics/shift_sales/;

-- join in SafeGraph data
CREATE OR REPLACE TABLE frostbyte_tasty_bytes_dev.harmonized.shift_sales
FROM frostbyte_tasty_bytes_dev.raw.shift_sales a
JOIN frostbyte_safegraph.public.frostbyte_tb_safegraph_s b
ON a.location_id = b.location_id;

-- promote the harmonized table to the analytics layer for data science development
SELECT * FROM frostbyte_tasty_bytes_dev.harmonized.shift_sales;

-- view shift sales data


Tasty Bytes is aiming to achieve 25% YoY sales growth over 5 years. To support this goal and maximize daily revenue across the fleet of trucks, the data science team needs to build an ML model to direct trucks to locations that are expected to have the highest sales on a given shift.

Step 1 - Cloning the GitHub Repository

Clone the GitHub repository. This repository contains all the code you will need to successfully complete this QuickStart Guide.

$ git clone

Step 2 - Updating the Authorization File

Step 3 - Creating the Python Environment

From the terminal, execute the following to build the Python environment and launch Jupyter Notebook:
i. Create a Python 3.8 environment called "py38_env_tb1" (tb1 = Tasty Bytes 1) using packages (& versions) from the Snowflake Anaconda channel

conda create --name py38_env_tb1 --override-channels -c python=3.8

ii. Activate the py38_env_tb1 environment

conda activate py38_env_tb1

iii. Install the Snowpark Python package and packages that will be used in functions deployed on Snowflake from the Snowflake Anaconda channel

conda install -c snowflake-snowpark-python numpy pandas scikit-learn joblib cachetools

v. Install packages that will be used only in the Python environment (i.e. UI, visualization..)

pip install streamlit matplotlib plotly notebook

vi. Navigate to the cloned GitHub repo and launch Jupyter Notebook

jupyter notebook

Step 4- Running the Jupyter Notebook



Now that you have deployed a model that predicts the shift sales of each location for the upcoming shift, you want to find a way for truck drivers to use these predictions to pick where they will park. You need to create an application prototype to show the engineering team how a truck driver would interact with the shift sales forecast model. The application will allow a user to pick a city and shift time (AM or PM) and show predicted sales by location on a map.

Step 1 - Confirming Prerequisites

Step 2 - Optional: Reviewing the Code

Step 3 - Launching the App

Overview of Objects Created in this Quickstart




Step 1 - Removing Snowflake Objects

USE ROLE accountadmin;
DROP PROCEDURE IF EXISTS, array, varchar, varchar);
DROP FUNCTION IF EXISTS, float, float, float, float, float, float, float);
DROP DATABASE IF EXISTS frostbyte_tasty_bytes_dev;
DROP DATABASE IF EXISTS frostbyte_safegraph;

Step 2 - Removing Python Environment

conda remove --name py38_env_tb1 --all

Step 3 - Removing Cloned GitHub Repository

rm -rf sfguide-tasty-bytes-snowpark-101-for-data-science


You did it! You have successfully complete the Tasty Bytes - Snowpark 101 for Data Science Quickstart.

By doing so you have now:

Next Steps

To continue your journey in the Snowflake Data Cloud, please visit the link below to see more Tasty Bytes - Quickstarts available to you.