Snowflake Connector | Structured Data and Graph Models

KumoRFM can connect directly to Snowflake data warehouses, enabling predictions on enterprise-scale data without moving it out of Snowflake.

Installation

The Snowflake backend requires the Snowflake connector:

$ pip install kumoai[snowflake]

Quick Start

From a Snowflake notebook (uses the active session automatically):

1 import kumoai.rfm as rfm
2 
3 graph = rfm.Graph.from_snowflake(database="MY_DATABASE", schema="MY_SCHEMA")

With explicit credentials:

1 graph = rfm.Graph.from_snowflake(
2     connection={
3         "account": "my_account",
4         "user": "my_user",
5         "password": "my_password",
6         "warehouse": "my_warehouse",
7     },
8     database="MY_DATABASE",
9     schema="MY_SCHEMA",
10 )

This will:

Connect to the Snowflake database and schema
Discover all tables in the schema
Infer column metadata (data types, semantic types, primary keys, time columns)
Detect foreign key relationships
Print a summary of the inferred metadata and links

Specifying Tables

Control which tables to include and customize their configuration:

1 graph = rfm.Graph.from_snowflake(
2     tables=[
3         "USERS",                                           # Include by name
4         dict(name="ORDERS", source_name="ORDERS_SNAPSHOT"),# Rename source
5         dict(name="ITEMS", schema="OTHER_SCHEMA"),         # Different schema
6     ],
7     database="DEFAULT_DB",
8     schema="DEFAULT_SCHEMA",
9 )

Table configuration options:

Key	Description	Required
`name`	The table name used in PQL queries	Yes
`source_name`	The actual table name in Snowflake (if different from `name`)	No
`database`	Override the default database for this table	No
`schema`	Override the default schema for this table	No
`primary_key`	Override the auto-detected primary key	No

Connection Options

There are several ways to establish a Snowflake connection:

1. Active session (Snowflake notebooks):

1 # No connection needed — uses the active Snowpark session
2 graph = rfm.Graph.from_snowflake(database="MY_DATABASE", schema="MY_SCHEMA")

2. Credentials dictionary:

1 graph = rfm.Graph.from_snowflake(connection={
2     "account": "my_account",
3     "user": "my_user",
4     "password": "my_password",
5     "warehouse": "my_warehouse",
6 })

3. Existing connection:

1 import snowflake.connector
2 conn = snowflake.connector.connect(...)
3 graph = rfm.Graph.from_snowflake(connection=conn, schema="MY_SCHEMA")

Database and Schema Defaults

The database and schema parameters set defaults for all tables. If not specified, the current database and schema from the active session are used.

Individual tables can override these defaults using the database and schema keys in their configuration dictionary.

Controlling Metadata Inference

1 graph = rfm.Graph.from_snowflake(
2     schema="MY_SCHEMA",
3     infer_metadata=False,  # Skip automatic type inference
4     verbose=False,         # Suppress output
5 )
6 
7 # Manually configure metadata afterwards:
8 graph.infer_metadata()
9 graph.infer_links()

Manual Edge Specification

1 graph = rfm.Graph.from_snowflake(
2     schema="MY_SCHEMA",
3     edges=[
4         ("ORDERS", "USER_ID", "USERS"),
5         ("ORDERS", "ITEM_ID", "ITEMS"),
6     ],
7 )

Supported Snowflake Types

KumoRFM maps Snowflake data types as follows:

Snowflake Type	KumoRFM Dtype	Default Stype
NUMBER, DECIMAL, INT, BIGINT, FLOAT, DOUBLE	`float` or `int`	numerical
VARCHAR, STRING, TEXT, CHAR	`string`	categorical / text
BOOLEAN	`bool`	categorical
DATE, TIMESTAMP, TIMESTAMP_NTZ, TIMESTAMP_TZ	`date`	timestamp
ARRAY	`stringlist`	multicategorical
VECTOR	`floatlist`	sequence