Can I use this skill for anomaly detection?

Yes. While not a dedicated anomaly detector, you can use the quantile forecasts as prediction intervals; values that fall outside the 90% confidence interval are flagged as statistically rare anomalies.

How does the preflight system checker work?

Before loading the model, the skill runs a script to verify available RAM, GPU memory, disk space, and Python versions to prevent system crashes during the inference process.

Do I need to train a model to use TimesFM Forecasting?

No, TimesFM is a zero-shot foundation model that works out of the box on any univariate time series data without the need for manual model training or parameter tuning.

What are the hardware requirements for this skill?

The recommended TimesFM 2.5 model requires at least 4 GB of RAM and 800 MB of disk space. While it works on CPUs, a GPU with 2 GB VRAM is recommended for faster inference.

Does it support external factors like holidays or price changes?

Yes, the skill supports exogenous variables through the 'forecast_with_covariates' function, allowing you to include dynamic numerical or categorical data.

TimesFM Forecasting

Name: TimesFM Forecasting
Author: Yezez9

byYezez9

0•

Data Science & ML

Performs zero-shot time series forecasting using Google’s TimesFM foundation model to predict trends across sales, sensors, and financial data without custom training.

The TimesFM Forecasting skill provides local, agent-friendly inference for Google Research's Time Series Foundation Model. It enables developers to generate point forecasts and calibrated quantile prediction intervals for any univariate time series data, ranging from a few data points up to 16,384 context points. By utilizing a pre-trained decoder-only architecture, it bypasses the need for manual model tuning or training, making it ideal for rapid demand forecasting, anomaly detection, and capacity planning. The skill includes a mandatory preflight system checker to ensure hardware compatibility before loading the model, ensuring safe local execution.

Key Features

010 GitHub stars

02Zero-shot forecasting for univariate data without requiring custom model training.

03Integrated preflight system checker for RAM, GPU, and disk space verification.

04Supports long-context windows up to 16,384 data points for complex historical analysis.

05Generates point forecasts and calibrated 10th-90th percentile quantile intervals.

06Native support for exogenous variables (covariates) like pricing and holiday flags.

Use Cases

01Monitoring sensor data to identify statistical anomalies using prediction intervals.

02Predicting retail sales or demand patterns using historical transactional data.

03Projecting energy consumption or resource utilization for infrastructure planning.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add yezez9/research-agent timesfm-forecasting

For use in Claude.ai and ChatGPT

Key Features

010 GitHub stars

02Zero-shot forecasting for univariate data without requiring custom model training.

03Integrated preflight system checker for RAM, GPU, and disk space verification.

04Supports long-context windows up to 16,384 data points for complex historical analysis.

05Generates point forecasts and calibrated 10th-90th percentile quantile intervals.

06Native support for exogenous variables (covariates) like pricing and holiday flags.

Use Cases

01Monitoring sensor data to identify statistical anomalies using prediction intervals.

02Predicting retail sales or demand patterns using historical transactional data.

03Projecting energy consumption or resource utilization for infrastructure planning.