batch_size: <list[int]> (Optional)

Supported Task Types

Default Values

Description

batch_size

Kumo.ai

Overview

Kumo Documentation

Quick Start

Data Connectors

Select Tables

Column Preprocessing

Create Graph

Explainability

Batch Predictions

Outputs

Kumo features can help you with Model Risk Management

Model Risk Management

Quotas and Limits

Data Processing Addendum

Browser

SSO Configuration Guide

Notebooks

Predictive Query

Column Processing

Encoder Overrides

SDK Reference

2025

Product Updates

2024

2024 Product Updates

2023

2023 Product Updates

Guide

Examples

Reference

Releases

Start for free

Community

Contact Support

AWS S3

Google Cloud BigQuery

Kumo offers three ways to connect to your Snowflake data warehouse. Click the tab below to learn more. - Snowflake Direct Connector - Snowflake Secure Data Sharing - Snowflake Native App

Snowflake

Databricks

Local Upload

Understanding Predictive Query

Predictive Query Structure

Task Types

Static vs. Temporal

Training

Run Mode

Model Plan

Model Plan Intuition

Evaluation

Classification

Link Prediction

Regression

Baselines

Deployment Modes

SaaS

Business Operations

Demand Forecasting Solution

Shipment Delay Prediction

Entity Resolution

Fraud

Chargeback Fraud Detection

Credit Card Fraud Detection Solution

Money Laundering Detection

Payback Abuse Detection Solution

Fraud Detection Demo

Growth and Marketing

Customer Churn Prediction

Lead Scoring Optimization

Customer Lifetime Value Prediction

Anti-Targeting Solution

Best Time to Send Notification

Personalization

Buy-It-Again Recommendation

Cold Start Recommendation

Item-to-Item Recommendation

Search Recommendations

Personalized Email Recommendations

Why Recommendation Systems are Better Off Using Hybrid Graph Neural Networks

Hybrid Graph Neural Networks

Try Something New

ASSUMING

CLASSIFY/RANK TOP K

FOR EACH

PREDICT

WHERE

COUNT

COUNT_DISTINCT

FIRST

LAST

LIST_DISTINCT

CONTAINS

ENDS WITH

IS NOT NULL

IS NULL

LIKE

NOT CONTAINS

STARTS WITH

activation

distance_measure

handle_new_target_entities

module

normalization

num_post_message_passing_layers

num_pre_message_passing_layers

ranking_embedding_loss_coeff

output_embedding_dim

prediction_time_encodings

target_embedding_mode

use_seq_id

num_neighbors

sample_from_entity_table

base_lr

early_stopping

lr_scheduler

loss

majority_sampling_ratio

max_epochs

max_steps_per_epoch

max_test_steps

max_val_steps

weight_decay

weight_mode

refit_full

refit_trainval

metrics

num_experiments

tune_metric

end_time

forecast_length

forecast_type

lag_timesteps

year_over_year

split

start_time

timeframe_step

train_end_offset

train_start_offset

What columns should I select in a table?

What Tables Should I Link?

How can I troubleshoot data quality issues?

How are Kumo table columns preprocessed?

General tips for ensuring data quality when creating graphs and executing pQueries and batch predictions.

How can I improve the quality of my data?

Preventing data leakage and handling time correctness

How does Kumo handle missing values in my dataset?

What mechanisms does Kumo provide to detect data leakage?

How does Kumo handle timezones for timestamp values?

What types of data can Kumo ingest?

How can I incorporate an external model for embedding a column?

How and when are timestamps used in Kumo?

How does kumo handle data governance and privacy

How should I scale/handle outliers in the data?

What model architectures does Kumo incorporate into its GNN design search space?

How do I use a feature store with Kumo?

How do I perform feature engineering with Kumo?

How do I specify the train/validation/test splits?

Best practices for selecting the right datasets to use in a predictive query.

Which datasets should I use for my predictive query?

Does a <MISSING> score field indicate that the column was excluded from training?

Can I use specific date and time values in my PQuery filters?

Debugging Model Performance

How do I improve model performance?

How do I perform backtesting on a holdout dataset?

Can I tune model hyperparameters?

How can I make my training jobs run faster?

How can I start with a smaller graph and/or a downsampled data set?

Learn how to use instance-level weights to influence model training in Kumo.

How do I assign different weights to training samples?

How does Kumo handle the cold start problem in ML?

How do I calibrate my model?

Best practices for comparing the performance of a Predictive Query against a historical baseline, heuristic, or machine learning model.

How can I compare a predictive query with an external model?

Why is my PQuery underperforming for a particular subset of data?

How do I generate predictions on new data using a previously trained model?

How can I diagnose problems with my data pipeline at batch prediction time?

What happens if the predictions per entity value in batch predictions differs from the initial pQuery’s RANK TOP K value?

What are Embedding Outputs?

Troubleshooting and Support

What ports does Kumo require access to?

run_mode	Default Values
FAST	`[512, 1024]`
NORMAL	`[512, 1024]`
BEST	`[512, 1024]`

Predictive Query

Model Plan

SDK

​batch_size: <list[int]> (Optional)

​Description

​Supported Task Types

​Default Values

`batch_size: <list[int]>` (Optional)

Description

Supported Task Types

Default Values