What are the data requirements for fine-tuning?

You generally need at least 100 observations per series, and data must be formatted with the Nixtla schema (unique_id, ds, y). This skill provides scripts to help prepare your data.

What is Nixtla TimeGPT?

Nixtla TimeGPT is a generative pre-trained transformer model specifically designed for time-series forecasting across diverse industries and frequencies.

How do I know if my fine-tuned model is better?

The skill includes a comparison feature that calculates error metrics like SMAPE for both the zero-shot and fine-tuned models on a held-out test set.

Why should I fine-tune a TimeGPT model?

While TimeGPT offers strong zero-shot performance, fine-tuning allows the model to adapt to the specific patterns, seasonalities, and noise present in your unique proprietary data.

Does this skill require a Nixtla API key?

Yes, you must have a NIXTLA_API_KEY set in your environment variables to submit training jobs and receive model IDs from the Nixtla API.

Nixtla TimeGPT Fine-Tuning Lab

Name: Nixtla TimeGPT Fine-Tuning Lab
Author: intent-solutions-io

byintent-solutions-io

データサイエンスとML

Optimizes Nixtla TimeGPT models through automated dataset preparation, job submission, and performance benchmarking for domain-specific forecasting.

概要

The Nixtla TimeGPT Fine-Tuning Lab skill empowers developers and data scientists to move beyond zero-shot forecasting by optimizing TimeGPT models on proprietary datasets. It automates the complex lifecycle of custom model training, including data validation against Nixtla schemas, asynchronous job submission, and progress monitoring. By providing built-in benchmarking tools to compare zero-shot performance against fine-tuned results, it ensures users can quantify accuracy improvements and deploy the most effective forecasting models for industries like retail, finance, and energy.

主な機能

Production-ready Python script generation for forecasting pipelines
Streamlined fine-tuning job submission and API management
0 GitHub stars
Real-time monitoring of training job status and completion
Automated dataset validation and Nixtla schema formatting
Performance benchmarking for zero-shot vs. fine-tuned models

ユースケース

Customizing energy grid load predictions through domain-specific model optimization
Improving retail demand forecasting accuracy using historical store-level sales data
Benchmarking the ROI of custom-trained time series models against base API versions

概要

主な機能

Production-ready Python script generation for forecasting pipelines
Streamlined fine-tuning job submission and API management
0 GitHub stars
Real-time monitoring of training job status and completion
Automated dataset validation and Nixtla schema formatting
Performance benchmarking for zero-shot vs. fine-tuned models

ユースケース

Customizing energy grid load predictions through domain-specific model optimization
Improving retail demand forecasting accuracy using historical store-level sales data
Benchmarking the ROI of custom-trained time series models against base API versions