https://www.nixtla.io logo
Join Slack
Powered by
Hi <@U064YF2VCDP> today we're releasing the new ve...
# timegpt
m
Hi @Vidar Ingason today we're releasing the new version of nixtlar (v.0.6.0), which uses the new API for TimeGPT. In a few hours it should be in main.
The issues you're encountering are probably related to that.
v
Ok no problem 🙂
e
Hi @Mariana Menchero after the change, will it be necessary to generate a new API key?
m
Hi, no, your API key should still work fine.
e
Hey @Mariana Menchero do you mean that once everything starts working again, my API should also start working? Or that it should already be working at this moment? Because since yesterday, when the predictions stopped working, they haven't resumed yet, and in the panel, it looks as if I don't have any API.
m
Ah, sorry @Ezdrasz Golaszewski we also changed the TimeGPT dashboard and I assumed you already had an API key there. Please create a new one and let us know if you encounter any issues.
e
@Mariana Menchero Letting you know 🙂 error500 after attempting to generate a new API:
m
Thanks for letting us know. We'll look into that asap
@fede (nixtla) (they/them)
f
hey @Ezdrasz Golaszewski! thank you for reporting. could you try again? 🙂
e
Hello @fede (nixtla) (they/them) 🙂 I can already see my API in the panel, but unfortunately my system hasn't been working since 07.10. I tried generating a new API, but it didn't work with that either. I checked with the following code: from nixtla import NixtlaClient api_key = "API" nixtla_client = NixtlaClient(api_key) if nixtla_client.validate_api_key(): print("Your API is OK!") else: print("Your API is not OK!") and I received a message that the API is incorrect. I uninstalled 'nixtla' as a test and installed 'nixtlats' and it started working, but with a warning: 'This package is deprecated, please install nixtla instead. Could something have changed?
Additionally, I see that when I run it on 'nixtlats', it gives different results for the same input data sets compared to 'nixtla', so it is important for me to be able to continue working with 'nixtla'.
I checked one more solution; I created an entirely new environment with a new Python installation and the most up-to-date libraries, and now everything has started working correctly. However, a new issue has arisen. 🫣 I quickly checked 5 results for the same data in three options: the option I ran before 07.10.2024, the option where I used the old 'nixtlats' for testing, and the 'nixtla' option where everything worked. Despite the differences between option 1 and 2 in terms of result values, you can see from the charts that the shapes are identical. However, the version that worked after 07.10.2024 shows completely different results. Did you update/retrain the entire TimeGPT, resulting in different outputs, or am I doing something wrong? Here’s a preview of the results: https://docs.google.com/spreadsheets/d/1ipsZz2FpDyEktz5KkIY9vfujrSRaQL8VFYOR2o13EC4/edit?usp=sharing For the past two months, I’ve been doing large-scale backtesting, and if TimeGPT has indeed been updated and is producing different predictions, I would need to redo all the backtests because we are fine-tuning the collaboration between TimeGPT and our financial model. I’ve mostly exhausted my query limit for this month. Would it be possible to purchase additional queries for this month at a promotional price so that I can rerun the tests?
j
So the input is B2:B6 and you're using the forecast method with h=12? It'd be easier for us if you could provide the code you're using to make the comparisons
👍 1
f
hey @Ezdrasz Golaszewski! absolutely, i already gave you more credits for this month. let us know if you need more.:) by the way, i resent you an invitation to our support channel, we can give you further assistance there. please feel free to share your code if possible. we'll be happy to help! 🙌
e
Hi @fede (nixtla) (they/them)! Thank you very much 😊, I have already uploaded it to the common channel
Open in Slack
PreviousNext
Powered by