Machine Learning & AI for better predictions

Hey there
I just found this awesome app yesterday and already bought Gold as I like all the possibilities and sources

Unfortunately it’s not so easy to compare different sources or to know which source is best for my place.

Have you ever thought about adding some kind of AI (like a neural network) to merge data of different sources to get even better predictions? Would be quite easy and could be done on the device or in the cloud, as oyu have all the needed data already.
Would be a great USP to have the best predictions for every specific location and I don’t know of any other tool that has this feature. And it might reduce complexity of the UI as users don’t have to care what service the need (except they want to).

Let me know if you’re interested and need More informations how to do this. Fortunately the needed libaries are all free.

Best
Big_Berny

2 Likes

@Big_Berny Hello and Welcome to the Forum and Thank You for using Flowx

1 Like

Hi @Big_Berny, and welcome to the forum!

There is a way to

Duble-tap on a graph, or long press and select compare:

https://flowx.io/help/basics/add-graph/

@Ohan Thanks, that’s useful to compare indeed. :slight_smile:
But for a “normal” user it’s quite an effort to compare the data manually.

That’s why an AI approach that generates better predictions based on the different sources would be much more powerful.

2 Likes

Hi! @Big_Berny!

I see and can agree with your point.

However, @duane normally replies that Flowx is an app for presenting available data and predictions made by others, and that he is staying away from making predictions himself.

Hi @Big_Berny,

I would love to do more AI stuff. There are few thoughts to consider:

  1. there is a merged regional model over the USA that I plan to add soon.
  2. different models are better for different location, e.g., GDPS tends to be better for higher latitudes compared to GFS. So merging these will average this out, i.e., it’ll be better for lower latitudes compared to GDPS and worse for higher latitudes compared to GDPS.
  3. if we merge the data, we’ll end up with something like MOS (Model-Output Statistics) which exists in all other typical weather apps.
  4. I agree AI is easy to do/apply but it’ll be damn hard to do well. Many weather orgs are doing it.

Feel free to post or email your ideas on how to do it. They might be different to my ideas.

BTW, I plan to add a dual side-by-side map mode to compare different models side by side.

Cheers, Duane.

4 Likes

Hi Duane
Very interesting.

To be honest I’m not experienced with weather data so I don’t know yet what’s the best approach. I would need to do some research first and play around with the data and some neural networks.
Would be fun but no idea how well it would work though. :slight_smile:

Do you have ground truth data, i.e. how the weather in the end actually was?

Best
Big_Berny

3 Likes

Coincidentally, I was emailing a user today who I send this link to a couple of years ago:
https://stackabuse.com/using-machine-learning-to-predict-the-weather-part-2/

You can see this Neural Net wasn’t much better than regression.

I’ve spent years collating forecast data, getting ground truth is the other half of the problem. And doing this globally and merging it together is a massive job. This is why I haven’t done it.

It may be best to start with one location. I think the link I posted pretty much guides you through this.

Cheers, Duane.

4 Likes