This Quick Start demonstrates how to use Kumo to predict customer churn (i.e., which active customers will stop buying) in just a few minutes.

Prerequisites

• Sign Up for a Free Trial: If you haven't already, sign up for a free trial at Kumo Free Trial.
• [SPCS Only]: Ensure the Kumo app is installed and started. If you haven't already installed or started Kumo, follow the instructions on how to install Kumo for SPCS.

What You Will Learn in This Quick Start

You will learn how to use Kumo's graph neural networks (GNNs) to predict customer churn. This Quick Start uses the H&M Personalized Fashion Recommendations dataset, which includes detailed information about customer transactions, articles, and customers. As you will see in this tutorial, the Kumo GNN is highly effective at learning the relational patterns between customers and articles, resulting in a model with extremely high accuracy.

The dataset is organized into three main tables:

• Customers: Contains information about the customers, such as customer ID, name and status.
• Articles: Contains information about the articles, such as article ID, name and category.
• Transactions: Contains transaction details, including transaction ID, customer ID, article ID, transaction date, quantity, and price.

Depending on your database, you can either directly go to Step 1 and add the data via an AWS S3 link to Kumo, or—if you are accessing Kumo via SPCS—you should start with uploading the data to your Snowflake account via Step 0

Step 0 [SPCS Only]: Upload Data to Your Snowflake Account

Step 1: Set Up Your Connector

Setting up the connector ensures that Kumo can access your data, enabling seamless data integration for model training.

1. Open Connector Configuration:

   • In the left-hand menu, click on Connectors.
   • Click Configure Connector to open the "New Connector" modal window.

2. Create a New Connector:

   i. Provide a name for your new connector (e.g., hnm_connector).

   ii. Choose your data source:

   • For Snowflake, enter the following details:

     • Account Identifier: ORGNAME-ACCOUNT_NAME. Find your account details.
     • Database: The database where your H&M data exists. Ensure your user has USAGE privileges.
     • Warehouse: The warehouse to process data. Ensure your user has USAGE privileges.
     • Schema Name: The schema to load tables from. Your user should have USAGE and SELECT privileges, and CREATE TABLE for writing back predictions.
     • User: Your username for Snowflake.
     • Password: Your password for Snowflake.

   • For AWS S3:

     • In the S3 Path textbox, enter: s3://kumo-public-datasets/Kaggle_H&M_reduced/parquet/.
     • Click the Validate button to see all files within the H&M public dataset directory.

   • For other data sources: See Data Connectors.

   iii. Click Done to save the connector.

Step 2: Register Table Schema

Now that you've set up your connector, it's time to register the H&M dataset within Kumo. This step defines how columns are later encoded for the model and sets the requirements for the graph schema registration by defining the primary keys of each table.

1. Navigate to Tables Overview:

   • In the left-hand menu, click on Tables.

2. Add Data Tables (Customers, Articles, Transactions):

   Follow the steps for each table:

   1. Click Add Table.

   2. Set the table name ( ARTICLES, CUSTOMERSor TRANSACTIONS )

   3. Select the appropriate connector (e.g., hnm_connector).

   4. Choose the respective table from your data source.

   5. Wait for column types to be inferred based on sample statistics. If this takes too long (this can happen in SPCS deployments when the warehouse you are using is busy), skip the inferring and set the column types manually. Follow these steps:

      1. Check the type of each column: The type of a column refers to how the data is later encoded. Kumo supports preprocessing for the following generic types:

         • Numerical: Integers and floats where the ordering of numbers from lower to higher values has semantic meaning (e.g., product price, percentage discount).
         • Categorical: Boolean values or string values typically only a single token in length, with limited semantic meaning, and up to 4000 unique values (e.g., premium subscription status).
         • Multi-categorical: Comma-separated variable length lists of categorical values (e.g., a list of product attributes or categories).
         • ID: Columns that correspond to primary or foreign keys in a relational schema. (e.g., customer ID, product group number).
         • Text: String values with multiple tokens in length, where the content has semantic meaning (e.g., product description).
         • Timestamp: String or format-specific date/timestamp values (preferably ISO 8601). Ensure date/time format is valid.
         • Embedding: Lists of floats, all of equal length; typically the output of another AI model.

      2. Check the data type: This refers to the underlying data type. Data type checking is especially relevant for connecting tables later, as only columns with the same type can be connected via a Primary/Foreign key link. No modification is needed for data types for the H&M dataset.

      3. Set the primary key: This is the unique identifier column of a table and is used later within the Primary/Foreign key connection.

      4. Set a create date: For tables that correspond to real-world events (such as transactions), this is the time that the even took place in the real world. Kumo uses this to automatically prevent data leakage during training time, which is a very common mistake when building models by hand.

   6. Verify that all properties are set correctly prior to saving:

      1. ARTICLES:

         

      2. CUSTOMERS (the warnings can be ignored):

         

      3. TRANSACTIONS:

         
      
      

   7. Click Save to save each table.

Within the Tables page, you should now see three tables:

Step 3: Create a Graph Schema

Now that your tables are registered, it's time to define the graph schema, which specifies how your data tables are connected. This step is crucial for using GNNs effectively.

1. Navigate to Graph Setup:

   • In the left-hand menu, click on Graphs.

2. Create a New Graph:

   i. Click Create Graph in the top right-hand corner.

   ii. Provide Graph Details:

   • On the "Graph Setup" page, provide a name for your new graph (e.g., hnm_graph).

   iii. Select Tables:

   • Select the TRANSACTIONS, CUSTOMERS, and ARTICLES tables by checking the respective boxes.

   iv. Configure Graph Links:

   • Click the Next button on the top right-hand corner.
   • Kumo will automatically detect and suggest linkages.
   • Verify the linkages:
     • Ensure CUSTOMER_ID in CUSTOMERS is linked to CUSTOMER_ID in TRANSACTIONS.
     • Ensure ARTICLE_ID in ARTICLES is linked to ARTICLE_ID in TRANSACTIONS.
   • Click the Confirm Group button for both column groups.

   v. Complete Graph Creation:

   • Click on the Complete Graph Creation button on the top right-hand corner.

With this, all preparation is done. You should see the following graph now:

Step 4: Train Your First Model via PQL

After specifying the graph, you can train a model using Kumo's predictive query language (PQL). PQL helps specify the training table, defining the target and the entity for which you want to make predictions. In this case, you want to predict whether each customer is likely to churn within the next 90 days, focusing on those who have made a transaction within the last 60 days.

1. Open Predictive Query Creation:

   • Click on Write Predictive Query in the top right-hand corner.

2. Train Model via a Predictive Query:

   i. Provide a name for your predictive query (e.g., hnm_churn).

   ii. Select the graph you created earlier from the drop-down menu.

   iii. Specify predictive query:

   • In the predictive query text area, insert the following PQL statement to predict whether a customer that has made a transaction within the last 60 days is likely to churn within the next 90 days:
     PQL

     PREDICT COUNT(TRANSACTIONS.*, 0, 90, days) = 0
     FOR EACH CUSTOMERS.CUSTOMER_ID WHERE COUNT(TRANSACTIONS.*, -60, 0, days) > 0
     

     The query defines how the training data is automatically generated. In this example, Kumo will split the timeline of transaction data into intervals of 90 days, training the model to predict each customer's future number of transactions, and whether they will be greater than 0 within the next 90 days based on their transaction history, training for each of the past timeframes. During evaluation and future predictions, the model will respond with the likelihood of no transactions, evaluating against the actual number of transactions.

   iv. Click Next to specify the Model Planner.

   • You can use the default settings for now; however, when training models on your own data later, you can change these settings to influence the model architecture used, the number of examples trained, and more.

   v. Click Save and Train.

   

Congratulations, you have started training your first model! This can take a few minutes as Kumo ingests the data, encodes the columns, builds the computation graph, trains the models, and evaluates the models. You can follow the process as Kumo will show training table statistics and metric curves such as loss curves.

When the training is complete, take a look at the results by:

1. Checking out the evaluation metrics within the evaluation tab.

2. Inspecting the holdout set either via the Download button or via the shown SQL command within Snowflake for SPCS deployments .

3. Explore example predictions in our explorer tab, which shows you the specific predictions made by the model and how they match the actual outcomes.

Conclusion

🚀 You have now successfully created and trained your first predictive model using Kumo. Here is a summary of what was covered in this Quick Start:

1. Set Up Your Connector: Connected your data source to Kumo.
2. Register Table Schema: Defined the table schema.
3. Create Your Graph: Defined the relationships between tables.
4. Create Your PQuery: Wrote and trained models via a predictive query to make predictions.

Next, you can start training models on your data. To learn how to write a predictive query, take a look at the Predictive Query Tutorial. Additionally, you can learn how to create batch predictions or how to productionize your model.