search

time-series-foundation-models / lag-llama

Lag-Llama: Towards Foundation Models for Probabilistic Time Series Forecasting
Apache License 2.0
1.21k stars 147 forks source link

Exogeneous variables #82

Open EgorSheluto opened 3 months ago

EgorSheluto commented 3 months ago

Hi! Do you support exogeneous variables for zero-shot predictions and training from scratch? As I see gluonts supports it like static and dynamic features but not sure this works for lag-llama too

ashok-arjun commented 2 months ago

Hi!

We didn't pretrain the model with support for covariates, so it can't be used zero-shot/finetuned with covariates. But an idea is you can pretrain it from scratch again, with covariate support if you'd like. We don't have code for this now too. Please check this issue for more updates: https://github.com/time-series-foundation-models/lag-llama/issues/58