Quickstart

Install the SDK

Install the OuterProduct SDK from PyPI.

pip install outerproduct

Or with the uv package manager:

uv add outerproduct

Initialize and open a workspace

Call op.init() once to authenticate, then open a Workspace. The workspace is your entry point for unified data queries, model training, and reasoning.

import outerproduct as op

op.init()  # reads OUTERPRODUCT_API_KEY from the environment to authenticate
ws = op.Workspace()

Register a source table using the appropriate register_* function. For example,register_csv registers a csv file to the workspace. (See the Overview for an example of registering multiple sources.)

ws.register_csv("customers.csv", table_name="customers")

Build a training set and fit a reasoning model

Query over the registered sources, then train a ReasoningModel. (See the Overview for an example of training over multiple models, setting hyper-parameter configurations, and more.)

train = ws.table("customers")
model = op.reasoning.fit(train, task=op.Binclass(label_column="churn")).wait()

Run inference with native reasoning

Reasoning models come with additional reasoning functions. For example, get predictions with real-time drivers. (See Overview for additional examples of reasoning like running scenarios.)

recent = ws.table("customers").filter(op.col("signup_year") == 2025)
predictions = model.predict(recent)
reasoning = model.explain(recent)

print(reasoning.feature_names)
print(reasoning.drivers.shape)  

Putting it all together

import outerproduct as op

op.init()  # reads OUTERPRODUCT_API_KEY from the environment
ws = op.Workspace()

# Register a CSV as a source table you can query.
ws.register_csv("customers.csv", table_name="customers")

# Build the training set as a query, then fit a ReasoningModel.
train = ws.table("customers")
model = op.reasoning.fit(train, task=op.Binclass(label_column="churn")).wait()

# Score recent customers with a query; get predictions and reasoning.
recent = ws.table("customers").filter(op.col("signup_year") == 2025)
predictions = model.predict(recent)
reasoning = model.explain(recent)

print(reasoning.feature_names)
print(reasoning.drivers.shape)

Getting Started

Concepts

Data Connectors

Putting it all together

​Putting it all together

Putting it all together