Computing at scale

Handling large datasets is a common challenge in time series forecasting. For example, when working with retail data, you may have to forecast sales for thousands of products across hundreds of stores. Similarly, when dealing with electricity consumption data, you may need to predict consumption for thousands of households across various regions.

Nixtla’s TimeGPT enables you to use several distributed computing frameworks to manage large datasets efficiently. TimeGPT currently supports Spark, Dask, and Ray through Fugue.

In this notebook, we will explain how to leverage these frameworks using TimeGPT.


  1. Getting Started

  2. Forecasting at Scale

  3. Important Considerations

Getting started

To use TimeGPT with any of the supported distributed computing frameworks, you first need an API Key, just as you would when not using any distributed computing.

Upon registration, you will receive an email asking you to confirm your signup. After confirming, you will receive access to your dashboard. There, underAPI Keys, you will find your API Key. Next, you need to integrate your API Key into your development workflow with the Nixtla SDK. For guidance on how to do this, please refer to the Setting Up Your Authentication Key tutorial.

Forecasting at Scale

Using TimeGPT with any of the supported distributed computing frameworks is straightforward, as TimeGPT will read a pandas DataFrame and then use the corresponding framework. Thus, the usage is almost identical to the non-distributed case.

  1. Instantiate a NixtlaClient class.
  2. Load your data as a pandas DataFrame.
  3. Initialize the distributed computing framework.
  4. Use any of the NixtlaClient class methods.
  5. Stop the distributed computing framework, if necessary.

These are the general steps that you will need to follow to use TimeGPT with any of the supported distributed computing frameworks. For a detailed explanation and a complete example, please refer to the guide for the specific framework linked above.


Parallelization in these frameworks is done along the various time series within your dataset. Therefore, it is essential that your dataset includes multiple time series, each with a unique id.

Important Considerations

When to Use a Distributed Computing Framework

Consider using a distributed computing framework if your dataset:

  • Consists of millions of observations over multiple time series.
  • Is too large to fit into the memory of a single machine.
  • Would be too slow to process on a single machine.

Choosing the Right Framework

When selecting a distributed computing framework, take into account your existing infrastructure and the skill set of your team. Although TimeGPT can be used with any of the supported frameworks with minimal code changes, choosing the right one should align with your specific needs and resources. This will ensure that you leverage the full potential of TimeGPT while handling large datasets efficiently.