TimeGPT Excel Add-in

📘

Access token required

The TimeGPT Excel Add-in requires an access token. Get your token on the Nixtla Dashboard.

Support

If you have questions or need support, please email [email protected].

How-to

Settings

If it is your first time using the add-in, a welcome screen will guide you to a settings screen to configure your access token.

To access the setting later, click the three dots in the top right and click Settings.

Data Requirements

  • All data inputs must exist in the same worksheet. The add-in does not support forecasting using multiple worksheets.

  • Time series and exogenous variables (optional) must exist in their own columns. Values that you want predicted can be left blank.

  • Headers are optional.

Example:

timestampvalueexogenous1exogenous2
12/1/2016 0:00726150771066
12/1/16 1:0065.85952867311
12/1/16 2:0059.995881267470
12/1/16 3:0050.695767664529
12/1/16 4:0052.585680462773
12/1/16 5:0065.055849163780
12/1/16 6:0080.46162268346
12/1/16 7:002006807076799
12/1/16 8:00200.637049081312
12/1/16 9:00155.477018780506
12/1/16 10:00150.916830480066
12/1/16 11:006678978745
12/1/16 12:006545677975
12/1/16 13:006481877031
12/1/16 14:006339073875
12/1/16 15:006313571576
12/1/16 16:006331670560

Forecasting

Once you’ve configured your token and formatted your input data then you’re all ready to forecast!

With the add-in open, configure the forecasting settings by selecting the column for each input.

  • Timestamps - The column of the timeseries timestamps

  • Values - The column of the timeseries values for each point in time

  • Exogenous Variable(s) - The exogenous variables for each point in time. This should also include forecasting horizon (fh) additional timestamps to calculate the future values.

  • Frequency - The frequency of the data

  • Frequency Horizon - The forecasting horizon. This represents the number of time steps into the future that the forecast should predict. This is usually the number of empty cells in the value column.

  • Finetune Steps (advanced) - The number of tuning steps used to train the large time model on the data. Set this value to 0 for zero-shot inference, i.e., to make predictions without any further model tuning.

When you’re ready, click Run Forecast to generate the predicted values. The add-in will generate a new worksheet with the prediced values and a chart depicting the forecast.