SNOW partners with Tecton

SNOW yesterday partnered with Tecton to bolster its ML user experience/workflow.

Combine this news with Snowpark for python and Streamlit acquisition (which helps python scripts to apps from what I understand).

I am guessing this all pushes SNOW closer to parity versus databricks in the realm of ML use cases?……

In a statement on Wednesday, Tecton announced a partnership with the data giant under which the former’s feature store, known for managing the complete lifecycle of machine learning (ML) features, as well as the open-source one from Feast will be integrated with the Snowflake Data Cloud. The move, as the companies explained, will give enterprise data scientists a fast yet simple way to build production-grade features for a broad range of operational ML use cases, starting from fraud detection and product recommendation to real-time price tracking.

The move comes as the latest step from Snowflake to strengthen its data science play – one of the six workloads its supports through the Data Cloud along with data lake, data warehouse, data engineering, data application and data sharing.

A description of a feature store:

“A feature store is at its core a data warehouse through which developers of AI models can share and reuse the artifacts that make up an AI model as well as an entire AI model that might need to be modified or further extended. In concept, feature store repositories play a similar role as a Git repository does in enabling developers to build applications more efficiently by sharing and reusing code.…


What is Feature Store and how it can be used with Snowflake?
How does Snowflake make revenue from this?

As an example: Let’s say you have a credit card transaction event,
and you need to PREDICT from your Machine Learning model (ML model) if this is Fraud or not.

You first need to “write some code to build a ML model” that can give you this prediction,
but first you need sources of data for transactions of this credit card and this user.
Let’s say the source of data is in Snowflake.

The current modern way to write some code to build a ML model is to use a Feature Store.

What’s a Feature Store? - a set of processes, tools, platforms, ways of doing things that ingest source data, define and compute features, build and deploy Machine Learning models to production.

What’s a Feature? - any input into ML model to make prediction.

Here are good primers

There are many implementations of Feature Store; see a list of more than a dozen Most of them are NOT open-source.

The announcement from Snowflake and Tecton were about the two Feature Store implementations, both written in Python: Open Source supported by Tecton (and others) Tecton (built by people who previously built Michelangelo Feature Store at Uber)

It takes few lines of python code to connect Tecton to Snowlfake data source. Then few more lines of python code and you now use Tecton to create the pipeline to orchestrate the computation of features to feed the model (for training model and serving model)

Where is the revenue for Snowflake?
Once the pipeline is in place you run the computation of features over and over and over thus you consume Snowflake computing time.

(That is my assumption, I don’t use Snowflake, I would appreciate if someone could validate this)

With this assumption, this is now a new way to generate revenue from running batches at different intervals for building ML models.

short info on the integration…

Great use-case on fraud detection model with sample code, easy to read to non techies I think