These kind of papers often talk the world, but often lack a proper baseline model. They only compare against very simple (naive forecast), or non tuned models. In my experience a gradient boosting model will probably solve 95% of your forecasting problems, and trying to get fancy with a transformer (or even just a simple neural net) is more trouble then it is worth.
See for instance the pytorch geometric [1] package, which is the main implementation in pytorch. They also link to some papers there that might explain you more.
I am in the power forecasting domain, where weather forecasts are one of the most important inputs. What I find surprising is that with all the papers and publications from google in the past years, there seems to be no way to get access to these forecasts! We've now evaluated numerous of the ai weather forecasting startups that are popping up everywhere and so far for all of them their claims fall flat on their face when you actually start comparing their quality in a production setting next to the HRES model from ECMWF.
The wave length and pulse length aren't the same thing though. The short pulsed lasers actually consists of light that have many different wave lengths in them!
Not really sure what you are looking for, but the easiest might be to just add lags of your input variables in the same linear model that you are using.
If you are looking for an actual timeseries method I would checkout either darts [0] or statsforecast [1]. They are currently the most mature timeseries packages.
I wonder if it would be possible to probe what a model is trained on by usage of prompts the reply to which can only be answered well with certain training data.
For instance if I have some body of text that can't be found elsewhere on the internet, if the reply of the model references the information in that text in some way you may be fairly certain it was used in training.
The hard part is probably finding such a body of text.
Have been self-hosting this on a raspberry pi for the last 2 years, and it has been completely painless. It just keeps chugging along without needing to take a look.
As a European I am a bit surprised by this comment. Resistant based light bulbs have been banned already for some years, and all the LED lighting you can get in stores just work? Don't really have to think about it. Also would not assume that a 220V grid would make this easier.
As someone who mainly uses pandas, what is the benefit of using DuckDB to write your queries over using pandas (or polars) to operate on the data. Is it that you can already subset the data without loading it into memory?
Next to being easier to run, I also have the feeling that images by them selves are also more interesting then generating text. It is much easier for us to look past the defaults, whereas for a text it has to be of a really high quality to be of notice (and even more if you want it to be interesting).
Have the feeling this is definitely more of an American problem. I'm from a European country and the last time I have had any sort of power outage is more then 10 years ago (can't even remember it to be honest). Even with the current surge in prices the production and transmission system is still reliable as ever.
True, but this entire loop happens within the model, if you would return an output at every intermediate step the model would be extremely slow.
My take on why they have build the output layer like it is, is that next to feeling more human, it also forces you to be a bit more thoughtfull with your requests, and thus spam the system less. In the end it is still really expensive to run these models..