Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!

Options

Dave Tweed's sorta around again

Hi, I've been away for quite a while. I've forgotten: is it chat, strategy or both (or something else) which is only readable by registered Azimuth members?


and then realized the thing to do was compare what I could see as a logged in member vs when not logged in; it appears that chat and strategy are both members-only topics.

Anyway, I haven't been around for a bit. I'm goint to start throwing in the odd comment on anything I feel I can usefully illuminate. However, I'm not going to get deeply involved in the El Nino stuff because this is stuff the John and the Azimuth project plan to talk about at the next NIPS conference, so it's important that people can have confidence when things start to be done that they will be completed. At the moment I'm a bit flaky (which isn't a state secret but I don't want to put on the completely open internet), and one thing I don't like is being flaky in terms of failing to finally deliver something I've said I'll do that someone else is depending upon. (This flakiness is currently more intrinsic than a "I haven't got the time" thing; I'm able to control this flakiness, but I don't want to risk expand the "area" in which I need to control it at the moment.)

• Options
1.

Welcome back, David! John may need something other than the El Nino stuff at NIPS anyway...

Comment Source:Welcome back, David! John may need something other than the El Nino stuff at NIPS anyway...
• Options
2.
edited July 2014

Welcome back! I think I'll concentrate on El Niño-related stuff at NIPS. However, what they might like most is some sort of neural network / machine learning approach predicting El Niños. I wish there were some routine "off the shelf" freeware I could use to:

1. use machine learning to predict the value of a time series at time $t$ given one or more time series up to time $t - \Delta$.

2. rate how "good" such a predictive method is.

Obviously there are tons of choices here, but I imagine lazy people have some standard things they do. Can we learn what these are and actually do them before December?

Anyway, David: pointers to useful information, like answers to these questions, would be great! But I'll avoid relying on you in a way that would make me grumpy if you flake out.

Comment Source:Welcome back! I think I'll concentrate on El Ni&ntilde;o-related stuff at NIPS. However, what they might like most is some sort of neural network / machine learning approach predicting El Ni&ntilde;os. I wish there were some routine "off the shelf" freeware I could use to: 1. use machine learning to predict the value of a time series at time $t$ given one or more time series up to time $t - \Delta$. 1. rate how "good" such a predictive method is. Obviously there are tons of choices here, but I imagine lazy people have some standard things they do. Can we learn what these are and actually do them before December? Anyway, David: pointers to useful information, like answers to these questions, would be great! But I'll avoid relying on you in a way that would make me grumpy if you flake out.
• Options
3.

If all you're interested in is predicting a univariate time series with a neural network, I'm sure there are plenty of canned R packages that will do that (e.g., the 'forecast' package turns up in a quick Google search). Not sure if the same exist for predicting the evolution of spatial fields.

Not sure if the NIPS people will be impressed by some routine "off the shelf" approach, though.

Comment Source:If all you're interested in is predicting a univariate time series with a neural network, I'm sure there are plenty of canned R packages that will do that (e.g., the 'forecast' package turns up in a quick Google search). Not sure if the same exist for predicting the evolution of spatial fields. Not sure if the NIPS people will be impressed by some routine "off the shelf" approach, though.
• Options
4.

I have to do something routine and unimpressive before I can do anything better. Anyway, I don't really expect to impress them with my knowledge of neural networks by December; I mainly plan to say "here's an interesting hard problem; here's some stuff people have done; do better!"

Comment Source:I have to do something routine and unimpressive before I can do anything better. Anyway, I don't really expect to impress them with my knowledge of neural networks by December; I mainly plan to say "here's an interesting hard problem; here's some stuff people have done; do better!"