StackOverflow2013

Note that there are some explanatory texts on larger screens.

plurals

PO
text
Body
copied!<p>I assume that when you say you're trying to predict a time series you mean that you have a series of data and you're trying to predict the next value. If this is the case, you only want one input node, because you only have one piece of data at a time to base the prediction off of (if I'm making an incorrect assumption and you're actually trying to make one classification based off of the entire time series, then you would want n input nodes). </p> <p>If you're using the term "generation" in the genetic algorithm sense, I don't really see a way to avoid calculating error every generation, as you'll need to evaluate the fitness of all members of the population in every generation in order to generate the next generation. However, because you talk about reaching the end of the sample after a certain number of generations, it sounds like you might actually be talking about running the neural net on successive points in the time series. It does seem like you would want some sort of error metric that takes into account the error at each point in the time series. The easiest thing to do would probably be to just sum them up as you go. Dividing by k at the end to get an average would probably be more intuitive (the average error on a given estimation is something we tend to think about more than summed error), but since k is a constant it shouldn't actually matter which one you use for your fitness function, as long as you're consistent.</p> <p>If my understanding of your problem thus far is correct, you're basically evaluating fitness of each candidate neural net in your population by running the entire time series of data through it and keeping track of the error. Since the whole thing is your fitness function, you need to run it for each candidate structure in each generation. Thus, this sample could be construed as your training set. In order to evaluate success more generally, you would need to have a different time series to use as a test set. </p> <p>Hope that helps! Let me know if I misunderstood your question or if any part of that is unclear.</p>

Querying!

Guidance

An individual column

Larger individual text columns get their own page to allow for proper reading.

SQuiL has stopped working due to an internal error.

If you are curious you may find further information in the browser console, which is accessible through the devtools (F12).

Reload