hi @Max (Nixtla), yes I finally got to the moment where I am starting to work on trying to reproduce the nixtla M5 forecast! The work is a bit more involved than expected. I recall that my initial goal is to reproduce the M5 forecast as shown in your colab: https://colab.research.google.com/drive/1pmp4rqiwiPL-ambxTrJGBiNMS-7vm3v6. Since the API is no longer available as you told me in the call we had some time ago, I am trying to reproduce this locally using your open source libraries. I have to use data and code found in different places of your repos and apply some patches, until now the work has been more or less straightforward. But now, I just ran into a know issue (https://github.com/Nixtla/nixtla/issues/15) so it seems like a good moment to pause, say hi here and possibly ask for help! 🙂 you can see the work I have done so far in this public gist (it is a somewhat "dirty" notebook with my notes): https://gist.github.com/pietroppeter/d36d5f40ab3bb5cd6879b713d352d3ba not sure if any of you can help me out there

Great having you here @Paige Bailey! You are officially number 45 in our small community.

just wanted to pass on my congrats to the nixtla team re: http://nixtla.io/transfer-learning/! very exciting stuff

:) thanks @Chris Gervais

Thanks for the support 🙂

Welcome @Valeriy!!

Hi all, good to be here, greetings to @Max (Nixtla), @fede and the Nixtla team

We saw your call to action to the conformal predictions, is there a torch or pointers to the technique?

We are also curious about your experience with the transferability of the MQN-HiTS

@Kin Gtz. Olivares yes this is the place to go for conformal prediction. https://github.com/valeman/awesome-conformal-prediction have not spun MQN-HiTS yet but certainly on my plan of things to do.

Thanks for the pointer, we will take a look

There is conformal prediction slack as well https://app.slack.com/client/T02S1RLPGJE a few authors of time series conformal papers there already. Plus forecasting channel, most welcome to join and collaborate.

I think there are plenty of opportunities for conformal predictions to shine with neural nets, to fix their over confidence.

Have you guys considered adding large-scale pre-trained transfer learning N-HiTs trained on large retail Dunhumby dataset? It would probably be more powerful than one trained on M-comps. I have a link here to the dataset here https://www.dunnhumby.com/source-files/

@Kin Gtz. Olivares BTW DataRobot seems to went all on conformal prediction for time series about a year ago https://docs.datarobot.com/en/docs/release/public-preview/mlops-preview/prediction-intervals-regression.html#the-trumpet-chart must have picked it up from one of my posts. DriverlessAI next.

Fresh of the press with good coverage of Nixtla, also on LI, Medium and Twitter https://valeman.medium.com/python-vs-r-for-time-series-forecasting-395390432598

Hi everyone, I hope you are good! I have a question about statsforecast auto_arima: The documentation shows that we can execute auto_arima like this:

fcst = StatsForecast(
    series_train, 
    models=[(auto_arima, 12)], 
    freq='M', 
    n_jobs=1
)

But what if I want to restrict the search space of the AR, MA orders? How can I limit the order of (S)AR and (S)MA to, say, p=1, q=1, P=1, Q=1? Thank you! (please let me know if I should be asking this in some other channel)

https://analyticsindiamag.com/facebook-prophets-existential-crisis/ prophet in the news again

👋 Hi everyone!

Hi Everyone! I have a statsforecast arima model, and would like to perform rolling h-step predictions. My model is trained on very large quantity of data, so I do not want to retrain the model as new data comes in - essentially I want to use the actual new observations to predict the next h values, without using those observation as part of training. How can that be done? The model.predict() only allows to predict h-steps after the end of the training set, and does not allow me to append any new observations without training on them before. Thank you!

Welcome @Ada Canaydin! Happy to see you here.

Hi everyone! I am using 'tsfeatures' to extract features. However, I am getting the following error:

--> 916         ts_features = pool.starmap(partial_get_feats, ts.groupby('unique_id'))
TypeError: 'str' object is not callable.

If you have any recommendations on how to tackle this problem, it is much appreciated! I also attached my df here and thank you in advance :)

<!channel>: Yesterday we released our new 👑HierarchicalForecast library. 🔥 Show some support by giving us a 🌟 https://lnkd.in/eiCSKaPi.  Now you can reconcile forecasts for hierarchical problems in a simple way using statistical approaches such as Bottom Up, Top Down, Middle Out, Minimum Trace, and Empirical Risk Minimization. Moreover, you can achieve state-of-the-art results. See Benchmarks: https://lnkd.in/e8p3ntbM 🚀🚀🚀 Hierarchical 👑 forecasting is important where time series data can be grouped or aggregated at various levels. Classical examples include aggregation of sales from product level to brand levels or geographical aggregations from zip code to country. Since these categories are nested within the larger group categories, the collection of time series is said to follow a hierarchical structure.   🔥🔥🔥

👋 Hi everyone!

Hi Everyone! I am trying to train the MQNHITS model on my monthly time-series data whose structure is similar to that of M4 monthly time-series data I tried all the notebooks which are available on GitHub and did all the required configurations(“H” to “M”) but none of them is giving results. the training is automatically stopping in between as there is no proper documentation available please help. here is the screenshot of my data.

👋 Hi everyone!

Hello, how are you doing? I'm trying to adapt my work to auto_arima from the current use of pmdarima that is way too slow for the work I'm doing and it seems to work faster indeed. However I'm trying to see how to calculate the aic value of the model and for that I need the order (p and q), is there anything available to get this data

Hey Javier, have you tried the colab examples?

Here is a AutoARIMA/ETS tutorial.