search

KimMeen / Time-LLM

[ICLR 2024] Official implementation of " 🦙 Time-LLM: Time Series Forecasting by Reprogramming Large Language Models"
https://arxiv.org/abs/2310.01728
Apache License 2.0
1.02k stars 179 forks source link

How to do future demand forecasting with better results? #96

Closed ManukondaGopalakrishnaReddy closed 3 weeks ago

ManukondaGopalakrishnaReddy commented 1 month ago
  1. I have been experimenting with different parameters, still results are being deviated. Any suggestions on how can we improve the performance?
  2. How can we use TimeLLM for future demand forecasting effectively?

Your support is appreciated, thanks in advance.

kwuking commented 1 month ago

Hi, what exactly do you mean by "future demand forecasting"? I'm not quite sure about the specific context. What kind of demand are you referring to?

ManukondaGopalakrishnaReddy commented 1 month ago

Thanks for your response, I wanted to work with Time-LLM on sales data where I've to forecast future demand of a product. Unable to utilise it effectively. Seeking your help/lead. For furthur information we can interact via mail - m.gopalkrishnareddy2@gmail.com. Thank you once again.

ManukondaGopalakrishnaReddy commented 1 month ago

Thanks for reaching out. I have a product sales data. Demand is a column actually. I wanted to forecast the demand of that particular product. Referring the pull request by dahouathi2, I was exploring that. What was mentioned there is to predict the test dataset values. But I wanted to do future forecasting i.e let's say coming 1 week demand. Request you to suggest me the steps to do it. I am unable to figure it out.

Coming to another point, how can I improve the performance of TimeLLM on my data?

Thank you once again.

On Wed, May 29, 2024, 12:21 PM MetaKing @.***> wrote:

Hi, what exactly do you mean by "future demand forecasting"? I'm not quite sure about the specific context. What kind of demand are you referring to?

— Reply to this email directly, view it on GitHub https://github.com/KimMeen/Time-LLM/issues/96#issuecomment-2136664816, or unsubscribe https://github.com/notifications/unsubscribe-auth/ATLFZQSTCETEN3F2TFU6A7LZEV3FNAVCNFSM6AAAAABINPBOZOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCMZWGY3DIOBRGY . You are receiving this because you authored the thread.Message ID: @.***>

kwuking commented 3 weeks ago

Enhancing predictive performance is a direction we've been exploring, which may require you to further fine-tune with relevant data in order to achieve better prediction results that align with the characteristics of this dataset.