Time Series Forecasting with Lag Llama
Using the Lag-Llama model and time series data, I predict overnight temperatures to determine when to bring my orange mum plant indoors before freezing temperatures in New York.
MAIN POINTS FROM TRANSCRIPT
- The Lag-Llama model predicts overnight low temperatures to protect an orange mum plant from freezing.
- The model uses hourly temperature data from New York, gathered from ACS Web services.
- Lag-Llama operates without pre-training, leveraging large-scale time series datasets for predictions.
- The forecasting process involves creating a lag estimator and predictor using the GluonTS library.
TAKEAWAYS
- The Lag-Llama model can forecast without pre-training, similar to LLMs generating text.
- Hourly temperature data is crucial for predicting the first frost in late November.
- Missing data in the time series is addressed through interpolation.
- The model's transformer architecture utilizes lag features for accurate predictions.