One that maybe <@U06HUECPYN5> or <@U06H2MDHK7H> kn...
# supportt
Tracy Teal
04/18/2024, 6:18 PMOne that maybe @Mariana Menchero or @Marco know the answer to, with someone who is trying to forecast a table that has multiple unique_ids, that might depend on each other. Text in thread.
ā
1
Tracy Teal
04/18/2024, 6:20 PM_Iām experimenting with your TimeGPT model, and Iām trying to forecast a table that has multiple unique_ids, that might depend on each other._
This is the sample data Iām working with:
The ds column is Month, and the y column is Activity.
_The main unique_id column is Currency Pair, but I also have the Account column that distinguishes the Activity._
To use the TimeGPT APIs, I tried to pivot the table on Account, and have something like this:
_df = df.pivot_table(index=['Month', 'Currency Pair'],_
columns='Account',
_values='Activity').reset_index()_
What can you recommend for my scenario where the value to forecast could depend on N other columns, and the model automatically infers the relationships/weights?
m
Marco
04/18/2024, 6:27 PMMy understanding is that TimeGPT, by default, analyzes and predicts all series concurrently. However, if he wants to use cash and receivables as features, then he must pass them in the
X_dfm
Mariana Menchero
04/18/2024, 6:29 PMI agree with Marco, and would also recommend doing some finetuning. This seems to be a case where using the user's data is important, instead of just relying in the zero-shot model.
š 1
t
Tracy Teal
04/18/2024, 6:30 PMThanks both! I'll respond with that information.
m
Marco
04/18/2024, 6:31 PMWe can link to tutorials on the website: exogenous features and multiple series and finetuning
š 1
t
Tracy Teal
04/22/2024, 8:21 PM@Marco @Mariana Menchero Luca Spolidoro asked this question, and had some follow up questions to my response. So, I suggested he join our Slack so he can ask there directly. He's going to do that, so you'll likely see the follow up to this question there.
š 2