Hi all! A user was asking about finding the optimi...
# supporty
Yibei
06/11/2025, 5:29 PMHi all! A user was asking about finding the optimized set of parameters for TimeGPT. I remember @Nikhil Gupta had worked on a function for this?
n
Nikhil Gupta
06/11/2025, 5:31 PMNot a function. Just a piece of code that does Cross Validation and Grid Search. Let me pull it up.
Nikhil Gupta
06/11/2025, 5:33 PM@Yibei - Check this https://github.com/ngupta23/timegpt_business_logic/blob/main/timegpt_business_logic/cv.py
I will give you access shortly.
Nikhil Gupta
06/11/2025, 5:34 PMYou should have received the invite.
y
Yibei
06/11/2025, 5:34 PMFantastic! Thanks @Nikhil Gupta this is exactly what I need
n
Nikhil Gupta
06/11/2025, 5:35 PMIt is just an experiment. It uses MLFlow, but we should create one without it.
Nikhil Gupta
06/11/2025, 5:35 PMThis is the main notebook
https://github.com/ngupta23/timegpt_business_logic/blob/main/analysis/mlflow_gridsearch.ipynb
y
Yibei
06/11/2025, 5:36 PMI’m wondering if we’ll be publishing it recently?
Yibei
06/11/2025, 5:36 PMIt’s going to be very useful
n
Nikhil Gupta
06/11/2025, 5:36 PMYes, we should clean it up and make a tutorial or a GridSearch Class in the client. I can do that when I get some time.
👍 1
h
Han Wang
06/12/2025, 5:25 PMWait, I think we should get rid of mlflow, and also there are other search algos, random, bayesian, i think we should release a feature for this and then publish
n
Nikhil Gupta
06/12/2025, 5:26 PMYes of course. This was just an experiment. There is no need for MLFlow. I can create classes for Random search, Grid search and Bayesian search, etc