Quality of forecasted values is bad for simple trends Hi folks, I used the Excel plugin to forecast a DAU measure with a clear exponential or logarithmic progression, and Timegpt-1 is producing a slightly downward trend instead (see attachment, the light blue is the prediction). Am I doing something wrong? This is not at all what I expected... I also attached the data to see if you can repro.

Hi @Luca Spolidoro! I'll try to debug the issue, could you share the .xlsx again? It isn't allowing me to open it because of permissions, maybe in .csv

Thanks Edu, here the CSV file.

awesome, thanks

@Eduardo López Amaya Just a heads up, I tested the data with my API setup and got a similar result using TimeGPT-1 Long Horizon

@Sergio Kuan thanks for that!

Hi @Eduardo López Amaya any news on the investigation? I noticed this "flatter than expected" forecast on other monotone-increasing series, see TimeGPT prediction VS a simple polynomial interpolation in the attachments.

Hi @Luca Spolidoro, We just released a new version of the plugin which supports

finetune_steps

, by using it this problem should be fixedstart excel or click on the

i

button at the top right and then

reload

Hi @Eduardo López Amaya What would be the solution if I use a similar dataset with a standard API query?

Hi @Sergio Kuan you should add the finetune_steps parameter to your request. Docs here https://docs.nixtla.io/reference/forecast_forecast_post

Got it, thanks @Eduardo López Amaya!