I know that order of giving data to nn is important, but I want to know how can find a good way too choose this order.
It's called ``active learning'', cf. http://active-learning.net/ This doesn't, of course, mean that you won't get stuck in local ``extrema'' of the information transfer, since the approach consists in maximizing the local information transfer.
No, active learning is for sequence ordering and similar jobs with sequences.
but I have a set of training data to give my neural network. I want to choose the best order to gain high performance, but thank u for your reply
Active learning isn't limited to sequence ordering-that's the point. It's a procedure that is defined by the selection of examples so as to maximize the information transfer-which is one measure of ``high performance'' and, in fact, can be generalized to the maximization of any other objective function, the idea being that one doesn't present all the examples at once.
Thank you.
Do you have any document that mentioned exactly what I want. (control of input order) I dont have time to read about active learning first:(
A good starting point may be this one: http://pub.uni-bielefeld.de/publication/2302013
``Active data selection in supervised and unsupervised learning''
The order of data is defined by the dimensions and classes in the data set. The dimensions are the attributes affecting the response variable and the classes are different categories of the response variable. Its necessary to have a idea as to what is the order to data set as it helps you in identifying the number of hidden units in the neural network. Vague number of hidden units will either lead to under fitting or over fitting. The number of hidden layer units should be less than input units and larger than output units.
order of neural network based on hidden units
#rows < 2/3 * #rows + #rows < 2* #rows
#rows: Number of rows
No. it is the number of tuples in your Data Set. more specifically your Training Data Set.
- Badges
- Science topic