Been playing a lot with a the "new" (not really new but version 2.0 of reactive extensions is really fast now, and is worth the time to learn it), with RX , WCF's (to actually stream results) and encog.
Basically I made a WCF's which sits on a server listens for traffic (data).
Once it receives data it output the train network data over an IObservable callback to multiple subscribers (or one).
Is it FAST? not really..by that i mean , I can't yet stream gig's of data per seconds , and it's really stable (though i'll have a look at Mark's way to send data over the wire as it seems he HAS a way to send lots of data fast (he posted a project from bitbucket , which looks great)...(Though i could use WCF pipes/MQMQ's, but my comps are all over the planets so i m using TCP..).
The details :
Fix connections, MT4 , Ninja, etc send the data to the WCF's which in turn will make a "query" :
Data => WCF => OnNext => Scan => Distincts => Merging =>Publish =>
WCF (Concurrent Dictionary of callbacks .. callBack.Foreach(x=>SendData(x)) , keeping it simple) to callback Subscribers the resulting network outputs.
I'll try to transfer my projects from bitbucket to github and put up a link.
All in all , Encog should start playing with RX, and it fits the "Neural" mind, ( http://channel9.msdn.com/posts/Charles/Erik-Meijer-Rx-in-15-Minutes/ && http://rxwiki.wikidot.com/ ) .
Rx also has a Mono implementation (which i ve been playing with on the mac..)
Training could be done totally async with pushing Data to a "neural brain" , which will push outputs to subscribers.(Right now, i m just having a neural relay , not a neural WCF trainer..)..
will post a link soonish.
Oh and this : http://arrayvisualizer.codeplex.com/ (I love this extension :P ) .