:speech_balloon: <https://app.intercom.com/a/inbox...
# support
i
💬 Conversation (#180) User started a conversation using a Messenger Workflow
✅ 1
t
This person is asking
Hello, can I use the "Playground" page to make a forecast with Exogenous Variables? If yes, what is the format of the .CSV that I need to upload?
I think the answer to this is that they can't use the playground for this right now? @Nikhil Gupta
n
I don't think that should be an issue. The csv should contain ds, y and the exogenous variables, them TimeGPT should be able to handle it. I have not tried it yet, but let me check and I will get back to you.
t
great, thanks!
n
It seems to be working but I am not sure how it is getting the future exogenous variable values. @azul (she/her) (nixtla) - What happens when the data has exogenous variables and the user does not provide the X_df argument? Does it consider the future exogenous variables to be 0 or does it forecast them first before forecasting y?
@Tracy Teal - I think we need to make some enhancements to this playground to completely support exogenous variables.
@azul (she/her) (nixtla) - never mind. I see various options detailed here: https://docs.nixtla.io/docs/tutorials-exogenous_variables
@Tracy Teal - to addresss the user's questions, currently, only option 3b is supported right now. https://docs.nixtla.io/docs/tutorials-exogenous_variables#3b-forecasting-electricity-prices-using-historic-exogenous-variables
3a, 3c and 3d are not supported currently
t
Thanks, so they can use historical exogenous variables, and to do that their csv should contain ds, y and the exogenous variables, them TimeGPT will be able to handle it?
n
Yes, exactly. The example in the tutorial has the data format - here.
t
perfect, thanks. I'll share that with the user
n
@Tracy Teal - New feature in playground on forecasting with exogenous variables (unreleased): https://github.com/Nixtla/yolotl/pull/148. Any feedback is welcome.
t
Thanks! I'll take a look.