Custom Python Packages in Snowflake Notebooks Using .whl Files

Read Time:4 Minute, 13 Second

Custom Python Packages in Snowflake Notebooks Using .whl Files: Snowflake Notebooks are becoming a powerful environment for combining data engineering, analytics, and AI workflows directly within the Snowflake ecosystem. However, one common limitation practitioners encounter is:

“What if the required Python package is not available by default?”

Unlike traditional Jupyter environments, Snowflake operates in a controlled execution environment, which means installing third-party libraries is not always straightforward.

In this article, we will explore:

How to install custom Python packages in Snowflake Notebooks
Why .whl files are the most reliable approach
A real-world use case demonstrating the value
Step-by-step implementation (production-ready)

The Challenge

When trying a simple installation like:

!pip install faker

You may encounter:

Package sources not available

This happens because:

The notebook does not have access to PyPI
No External Access Integration (EAI) is configured
No Artifact Repository is attached

In enterprise environments, this is very common due to security restrictions.

The Solution: Installing Packages via .whl Files

The most reliable and controlled approach is:

Download the package locally → Upload to Snowflake Workspace → Install from .whl

This method:

Works without internet access
Ensures reproducibility
Fits enterprise governance models

Step-by-Step Implementation

Step-by-Step Implementation

Step 1: Download the package locally

On your local machine:

pip download phonenumbers

This will generate a .whl file like:

phonenumbers-8.13.XX-py2.py3-none-any.whl

Step2: Upload the .whl file into Snowflake Workspace

Open Snowsight
Navigate to Projects → Notebook
Go to Workspace Files panel
Click Upload
Upload the .whl file

Step 3: Install the package inside the notebook

!pip install phonenumbers-9.0.31-py2.py3-none-any.whl

Step 4: Import and use the package

import phonenumbers

print(phonenumbers.__version__)

Real-World Use Case: Phone Number Standardization

Customer data often contains inconsistent phone formats:

9876543210
+91-98 7654 3210
(415) 555-2671

This leads to:

Failed SMS delivery
Poor data quality
Integration issues with CRM systems

Solution using phonenumbers

Solution using phonenumbers

In this step, we use the uploaded third-party package phonenumbers to clean, validate, and standardize customer phone numbers.

The package was installed in Snowflake Notebook using the uploaded .whl file. Without that .whl package, this country-aware phone validation logic would not be available in the notebook.

What this code is doing

First, we define a simple country_map:

country_map = { "IN": "IN", "US": "US", "ES": "ES" }

This tells the phonenumbers package which country rules to apply while parsing a phone number.

For example:

IN applies Indian phone number rules
US applies United States phone number rules
ES applies Spain phone number rules

Parsing the phone number

parsed = phonenumbers.parse(phone_raw, country_map.get(country_code, "IN"))

This line converts the raw phone number into a structured phone number object.

For example:

9876543210
+91-98 7654 3210
(415) 555-2671

These are messy formats, but phonenumbers.parse() understands them based on the country code.

If the country code is missing or not found in our map, we default to India:

country_map.get(country_code, "IN")

Checking whether the number is valid

is_valid = phonenumbers.is_valid_number(parsed)

is_possible = phonenumbers.is_possible_number(parsed)

Formatting the number

If the phone number is valid, we generate two standardized formats:

“phone_e164”: phonenumbers.format_number(parsed, phonenumbers.PhoneNumberFormat.E164)

E.164 format is the global standard format used by APIs, SMS platforms, and CRM systems.

Example:

+919876543210

Then we also create a human-readable international format:

“phone_international”: phonenumbers.format_number(parsed, phonenumbers.PhoneNumberFormat.INTERNATIONAL)

Example:

+91 98765 43210

Handling invalid numbers

If the number is not valid, we return:

{ "phone_e164": None, "phone_international": None, "is_valid": False, "is_possible": is_possible }

Applying the function to the full DataFrame

Applying the function to the full DataFrame

result = df.apply( lambda row: standardize_phone(row["phone_raw"], row["country"]), axis=1 )

This applies the phone validation logic row by row.

Then we convert the result dictionary into a DataFrame and join it back with the original data:

result_df = pd.concat([df, pd.DataFrame(result.tolist())], axis=1)

So the final output contains both:

original raw phone number
standardized phone number
validation flags

Loading Results into Snowflake

What benefit did we get from the .whl file?

The key benefit is that the .whl file allowed us to bring an external Python package into Snowflake Notebook even when direct pip install from PyPI was not available.

Without the uploaded .whl file, Snowflake Notebook would not know the phonenumbers package.

Because we installed the .whl, we were able to: