> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.nvidia.com/sdgm/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.nvidia.com/sdgm/_mcp/server.

# Quickstart: Train Your First Model

#### Coming soon

**Kumo containers will be publicly available in the NVIDIA NGC Registry soon. Stay tuned.**

**Tip**

This page helps you set up the fully managed cloud version of Kumo. To start using the Kumo RDL SDK, follow the [instructions here](/fine-tuning/introduction).

In this guide, we walk you through an example to train a model to predict churn using the [H\&M Kaggle dataset](https://www.kaggle.com/competitions/h-and-m-personalized-fashion-recommendations/data). The data is pre-loaded in an Amazon S3 bucket. You'll learn how to connect data sources, choose tables, create a graph schema, and train a model.

## Step 1: Select Group and Project

When you first sign in, select a Project to open your workspace. If you belong to more than one Group, select a Group first, then choose a Project within it. Your selected Group and Project scope the data, assets, and work available during the session.

## Step 2: Connect data

Kumo integrates with a variety of data sources to ingest data for training and running models regularly.
For this example, we will use the H\&M dataset pre-loaded in an S3 bucket.

* Select `Connectors` from the side navigation menu (bottom left of your screen).

* Choose the `Amazon S3` option. Enter the S3 path: `s3://kumo-public-datasets/quickstart/` and add the other details to create a new connector.

## Step 3: Choose tables

We will now select and configure tables for model training.

* Navigate to `Tables` from the side navigation menu and select `Connect Table`.

* Select the `customers` table. Kumo will process the data and automatically infer data types and semantic types for each column. You can update from the dropdown as needed.

* Select the property for columns - assign a primary key and create/end date for temporal data.

Repeat the steps for the `transactions` and `articles` tables.

## Step 4: Create graph schema

Now that your tables are selected, it's time to link them to create a graph schema.

* Navigate to `Graphs` and select the tables we just created.

* You can preview the graph on the right as the tables are selected.

* Kumo will link the tables based on primary and foreign key names. You can update these links as needed in the next step.

We are now ready to train our model.

## Step 5: Train model

* From the side menu, click `New` > `Training`.

* Select the graph you created in the last step.

* Enter the Predictive Query. This query predicts the likelihood of no transaction (churn) over the next 90 days, for customers who completed any number of transactions in the last 60 days.

**`PQL`**

```Text PQL
PREDICT COUNT(transactions.*, 0, 90, days) = 0
FOR EACH customers.customer_id
WHERE COUNT(transactions.*, -60, 0, days) > 0
```

* Click `Start Training`.

Training takes a few minutes. Once complete, you can:

* Review the Evaluation metrics.

* Get Explanations for the overall model or individual predictions.

* [Run the model](/batch-prediction) for Batch Predictions or embeddings with new data.

## Step 6: Connect your own data

The next step is to train models with your own data in Kumo. You can select from a variety of other [data connectors](/data-connectors). To use Kumo in your own data warehouse, see the [available deployment options](/deployment-modes).