TimeGPT-1: production ready pre-trained Time Series Foundation Model for forecasting and anomaly detection. Generative pretrained transformer for time series trained on over 100B data points. It's capable of accurately predicting various domains such as retail, electricity, finance, and IoT with just a few lines of code 🚀.
These are feature requests from our users, compiled through individual meetings, with the numbers indicating how many times each feature was mentioned:
Forecast Model
[ ] Improve performance in time series with clear cyclicity or seasonality, compared against models like LSTM and XGB (4).
[ ] Multivariate time series modeling (3).
[ ] Enhanced accuracy for industry-specific datasets (2).
[ ] Improved handling of nonlinear forecasts.(1)
Exogenous Variables
[ ] Capability to input only historical exogenous variables or forecast them based on text, rules, or specific indications (3)
[ ] Automatic selection of relevant exogenous variables (2).
[ ] Enhanced interpretability with feature importance insights (2).
[ ] Support for exogenous variables with different frequencies (e.g., hourly, daily, weekly) (1).
Finetune
[ ] Greater control over finetuning options, allowing deeper customization based on specific rules (6).
[ ] Ability to persist or reset the model post-finetuning (4).
[ ] Option to initiate finetuning through text prompts (1).
[ ] Indicator to show performance improvement after finetuning (1).
[ ] Control over how many layers to finetune (1).
[ ] Automatic selection of the best parameters for the task (1).
[ ] Lower data requirements for minimum data points needed (1).
[ ] Assistance with choosing the optimal loss function (1).
Description
These are feature requests from our users, compiled through individual meetings, with the numbers indicating how many times each feature was mentioned:
Forecast Model
Exogenous Variables
Finetune
Use case
No response