Explore TDgpt
TDgpt is an external time-series data analysis agent that is compatible with TDengine's main process taosd, enabling seamless integration of time-series data analysis services into TDengine's query execution process. TDgpt is a stateless platform that includes the classic statistical analysis model library Statsmodel, embeds machine/deep learning framework libraries such as torch/Keras, and on cloud services, TDgpt has built-in support for TaoData's self-developed time-series foundation models (TDtsfm) and TimeMoE time-series foundation model.
The public database Time Series Prediction Analysis Dataset on the cloud service has prepared three datasets (login to the TDengine Cloud, click DB Mart on the lower left of the screen) to experience the powerful prediction and generalization capabilities of the time-series foundation models. Both time-series models are trained on these datasets. The built-in datasets are stored in the forecast database, which includes three data tables.
- electricity_demand_sub: Electricity demand data (electricity_demands) from different time periods.
- exchange_2_cpc_results_sub: Public dataset cpc from NAB.
- exchange_2_cpm_results_sub: Public dataset cpm from NAB.
TDgpt integrates the prediction capabilities of time-series foundation models, allowing you to easily invoke these capabilities using SQL statements.
-
You can invoke TaoData's time-series foundation model to predict data using the following SQL statement:
select _FROWTS, forecast(val, 'algo=tdtsfm_1,start=1324915200000,rows=300') from forecast.electricity_demand_sub; -
You can invoke the TimeMoE time-series foundation model to predict data using the following SQL:
select _FROWTS, forecast(val, 'algo=timemoe-fc,start=1324915200000,rows=300') from forecast.electricity_demand_sub; -
After the query is complete, you can display the data graph by clicking the Draw button in the Chart section.
Users can execute show anodes full; to query the list of pre-configured algorithms in the current system. For usage methods of each algorithm, please refer to the TDgpt Document。
