When should I use fine-tuning instead of standard TimeGPT?

Use this skill when TimeGPT's zero-shot performance is insufficient or when your data has unique, domain-specific patterns that the base model hasn't encountered.

What data format is required for fine-tuning?

Your dataset should be a CSV containing 'unique_id' for series identifiers, 'ds' for timestamps, and 'y' for the target numeric values.

What metrics are used to evaluate the fine-tuned model?

The lab generates a metrics.json file containing Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) calculated against your validation split.

Do I need a Nixtla API key to use this skill?

Yes, you must export your NIXTLA_TIMEGPT_API_KEY to your environment to authenticate the fine-tuning process.

Is there a minimum amount of data required?

For effective fine-tuning, it is recommended to provide at least 50 time series points per unique series to ensure the model can capture relevant patterns.

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

Fine-tunes the TimeGPT model on custom time series data to maximize forecasting precision for domain-specific applications.

概要

The Nixtla TimeGPT Fine-Tuning Lab enables developers to move beyond zero-shot forecasting by adapting the pre-trained TimeGPT model to their specific datasets. It provides a structured workflow for data preprocessing, training configuration, model execution, and performance evaluation. This skill is essential when dealing with specialized domains where general patterns are insufficient, allowing for higher accuracy in areas like retail demand, energy consumption, and financial forecasting by generating fine-tuned model weights and comprehensive MAE/RMSE metrics.

主な機能

0 GitHub stars
End-to-end fine-tuning execution for custom model adaptation
Automated data preparation for Nixtla-compliant time series formats
Configurable training parameters including learning rate, epochs, and batch size
Comprehensive evaluation metrics including MAE and RMSE calculations
Visual data verification and automated training/validation split handling

ユースケース

Adapting global forecasting models to specific retail store inventory levels
Improving energy load forecasting for smart grid management using local consumption data
Refining financial market predictions using historical proprietary data

概要

主な機能

0 GitHub stars
End-to-end fine-tuning execution for custom model adaptation
Automated data preparation for Nixtla-compliant time series formats
Configurable training parameters including learning rate, epochs, and batch size
Comprehensive evaluation metrics including MAE and RMSE calculations
Visual data verification and automated training/validation split handling

ユースケース

Adapting global forecasting models to specific retail store inventory levels
Improving energy load forecasting for smart grid management using local consumption data
Refining financial market predictions using historical proprietary data