Nixtla
Nixtla-1: production ready pre-trained Time Series Foundation Model for forecasting and anomaly detection. Generative pretrained transformer for time series trained on over 100B data points. It's capable of accurately predicting various domains such as retail, electricity, finance, and IoT with ju
Last verified:
What is Nixtla?
Nixtla is the first foundation model for time series forecasting and anomaly detection, providing state-of-the-art forecasting capabilities to help organizations make better decisions with their time series data. It is a production-ready generative pretrained transformer trained on over 100 billion data points across diverse domains including finance, retail, healthcare, energy, and IoT sensor data.
Key features include zero-shot forecasting (no training required), anomaly detection for historical and real-time data, support for exogenous variables, cross-validation capabilities, prediction intervals for uncertainty quantification, fine-tuning options to customize the model for specific datasets, and the ability to forecast multiple time series simultaneously. Nixtla works with pandas DataFrames and distributed computing frameworks like Spark, Ray, and Dask for scaling to millions of series.
Nixtla is designed for data scientists, analysts, and engineering teams who need accurate forecasting without building and maintaining complex models. It serves industries including retail (demand forecasting), energy (grid operations and trading), financial markets (price and volume forecasting), supply chain (inventory optimization), and IoT (sensor predictions). The Python SDK enables users to generate forecasts in just a few lines of code.
Nixtla pricing
Pricing model: Freemium
TimeGPT offers a 30-day free trial with no credit card required when you create an account. The trial access expires after 30 days unless upgraded to a paid Enterprise plan. Enterprise subscription plans are customizable and include scalable API calls, flexible user access management, and high-level support options (email, chat, phone, or dedicated support). Three usage options are available: API (no infrastructure management needed), Cloud Deployment on Azure (managed by Nixtla with quick setup and automatic updates), and Self-Hosted Deployment (host on your own infrastructure for greater data control). Custom plan options including API call limits, user seats, and support levels can be arranged by contacting [email protected]. Open source libraries (StatsForecast, MLForecast, NeuralForecast, HierarchicalForecast) are available free forever.
Nixtla pros
- State-of-the-art forecasting accuracy with zero-shot learning
- First foundation model specifically designed for time series data
- No training required - produces results instantly
- Works with just one observation per series
- Supports multiple time series forecasting simultaneously
- Includes anomaly detection for historical and real-time data
- Simple Python SDK with just a few lines of code needed
- Generates forecasts in seconds for production use
- Supports exogenous variables and calendar dates
- Provides prediction intervals for uncertainty quantification
- Fine-tuning available to customize model for your data
- Works with distributed frameworks (Spark, Ray, Dask) for large datasets
- Self-hosted deployment option for complete data control
- Produces stable and consistent predictions for identical inputs
- Long-horizon model available for extended forecast periods
- Cross-validation capabilities for model evaluation
- Save and reuse fine-tuned model parameters
- Built-in plotting method for interactive visualization
Nixtla cons
- Cannot handle missing values or irregular timestamps
- No support for Polars DataFrame format
- Forecast accuracy decreases as horizon increases beyond seasonal length
- Requires API key and internet connection for cloud API usage
- Free trial expires after 30 days without credit card
- Some model architecture details remain confidential
- Enterprise pricing requires contacting sales (no公开 pricing)
- WriteTimeout errors can occur with very large datasets
Frequently asked questions about Nixtla
What is TimeGPT?
TimeGPT is the first foundation model for time series forecasting. It produces accurate forecasts for new time series across diverse domains using only historical values as inputs. The model reads time series data sequentially from left to right, examining windows of past data as tokens and predicting what comes next based on identified patterns. Beyond forecasting, TimeGPT supports what-if scenarios and anomaly detection.
Is TimeGPT based on a Large Language Model (LLM)?
No, TimeGPT is not based on any large language model. While it follows the principle of training a large transformer model on a vast dataset, its architecture specifically handles time series data and minimizes forecasting errors, not text processing.
How do I get started with TimeGPT?
To get started, register for an account at the Nixtla dashboard. After confirming your signup via email, access your dashboard to find your API key under 'API Keys'. Then run 'pip install nixtla' to install the Python SDK, and initialize the NixtlaClient with your API key to start making forecasts.
How can I use TimeGPT?
You can use TimeGPT through the Python SDK or REST API. Both methods require an API key obtained upon registration. The Python SDK uses the NixtlaClient class with methods like forecast() and plot(). The REST API accepts POST requests to https://api.nixtla.io/timegpt with JSON data containing your time series.
Can TimeGPT handle multiple time series?
Yes, TimeGPT can forecast multiple time series simultaneously. You need to include a 'unique_id' column in your DataFrame to identify different series. The model learns across all series at once, which can improve individual forecasts through cross-learning.
Does TimeGPT support forecasting with exogenous variables?
Yes, TimeGPT can incorporate external variables into forecasts. You pass exogenous variables using the X_df parameter in the forecast method. This includes numeric features, categorical variables, and calendar dates/holidays through separate tutorials.
Can TimeGPT be used for anomaly detection?
Yes, TimeGPT includes anomaly detection capabilities for both historical data and real-time monitoring. You can detect anomalies using the detect_anomalies() method in the Python SDK, which identifies unusual patterns in your time series data.
What is fine-tuning in TimeGPT?
Fine-tuning improves TimeGPT's performance for your specific data patterns. During fine-tuning, the pre-trained model undergoes additional training using your dataset. You specify finetune_steps parameter in the forecast method. You don't need to fine-tune every series individually - the model fine-tunes across all series simultaneously. You can also save and reuse fine-tuned parameters.
Can TimeGPT handle large datasets?
Yes, TimeGPT works with distributed computing frameworks for large datasets with hundreds of thousands or millions of time series. It supports Spark, Ray, and Dask backends. For large datasets, increase the num_partitions parameter or use a distributed backend to avoid WriteTimeout errors.
What is the maximum forecast horizon allowed by TimeGPT?
TimeGPT has no maximum forecast horizon, but performance decreases as the horizon increases. When the horizon exceeds the data's seasonal length (e.g., more than 12 months for monthly data), you receive a warning about potentially less accurate forecasts. For best results, keep the forecast horizon within the seasonal pattern of your data. A long-horizon model (timegpt-1-long-horizon) is available for extended forecasts.