How to use RNN?

[muerthe_svk]

Anobody knows how to use that thing?

[GospodinNoob]

Hi.

Connect the "memory" sockets inside each RNN and stack them into a sequence

[muerthe_svk]

yeah I found out from some russian guy video how to do it. It just doesnt make sense. The design is not properly explained and tbh the arrows you need to connect are just useless design. There is only one way to connect them you can not modify anything so I dont understand why you designed this element like that.

for those who still dont understand how to use RNN here is screenshot

https://imgur.com/a/wftfDNV

[IsopropylPod]

Yeah, with the nodes being exposed like that I expected I could link it up to other modules or maybe some custom memory modules to increase learnning or store information longer, nope.

[p4rav0id]

Yes thats just stupid.the end is really boring, you just link several of them (after training) and thats all.

[GospodinNoob]

Originally posted by p4rav0id:

Yes thats just stupid.the end is really boring, you just link several of them (after training) and thats all.

Thanks for the feedback

[GospodinNoob]

Originally posted by Ghostbird:

The RNN really makes no sense at all. You have to connect the bottom arrows in a fixed manner, which means it forces the player to execute two absolutely trivial actions.

Furthermore, the accuracy number on the RNN is incomprehensible. On a Perceptron a 9% seems to indicate that it will erroneously output 9% of the input packets.

However I have to stack four 10% RNNs to get above 70% accuracy.

I'd expect four stacked errors of 10% to yield an error of 0.01%

Hi

As in real life, RNN predicts some first elements with a big error (because their memory is empty). So accuracy with ARMA and RNN can be some lower than predicted.