Matlab - how to explain the difference between the number of neurons in the output layer and the number of outputs of the artificial neural network (see attachment)?
I would expect that the number of outputs and the number of neurons in the output layer are always the same.
The number of neurons in input layer corresponds to the number of attributes, however the number of neurons in output layer is the number of classes.
Ok, but there are 712 neurons in the output layer and 1024 output values. The ANN structure is 496-1024-712-1024. I would rather expect 496-1024-1024. How to explain "712 (neurons) - 1024 (output values)"? How does it work?
Grzegorz,
What do you like to achive by using ANN? Do you predict sth or classifying? As ANN is one of machine learning tools, you define the output (in most cases) to train the net (to classify cases to the classes defined or to predict values defined). Based on the experience of ANN users, it can be said that accuracy of ANN predictions is much higher if only one value is predicted. The discussion of error is much more meaningfull then too. The key issue to clarify - predicting or classifying?
best regards
Hubert
In most cases, the number of output neurons and number of output values are same. And also number of classes and number of output neurons are same
Hi Grzegorz Kłosowski you are using a feed foward network with one hidden layer.This topology is probably the most use in regression.
As R. Senthilkumar mentioned in this topology is usual that output layer matches with the number of outputs in Matlab.
I ask you to guide us through your network creation process. First I think that will be useful to know if you used an app of Matlab GUI or a script.
In case you use an script will be useful to add it to the question. Specially the creation part.
All the best
The number of neurons in input layer corresponds to the number of inputs, the number of neurons in output layer is the number of classes for classification task or just one for regression task.
… Read more
Hi Grzegorz Kłosowski
I think the reason is because of the data. For being sure try top skip the configuration process and to go straight to training.
This way software adjuts size during training according to training data.
You can adapt and try any of the adjunct scripts.
A mysterious thing. I reduced the number of cases in the training set to 10 (to simplify the calculations) and I ran the script from Roberto Eloy Hernández Regalado. And now the ANN looks like attached.
Use this function : https://www.mathworks.com/help/deeplearning/ref/activations.html;jsessionid=93478d205259c84efb8fb905d436
I hope it is applicable to ANN too
Grzegorz Kłosowski
Your target data is only 0 and 1? If the answer is positive I think you should try a different kind of ANN like a neural cluster or a neural pattern recognition.
You can look for the Iris Flowers problem. I think this is a good example.
Also is good to know if all the cases are complete. Like I said before try nnstart command. This app helps a lot during the creation process doing the necessary transformations to the data.
Best regards
Hi Roberto Eloy Hernández Regalado,
Yes, my outputs are binary (0 or 1). However, my question does not concern a specific problem that I have to solve. I just can not understand how Matlab works. I showed that in some cases, internal Matlab algorithms automatically set (reduce) the number of neurons in the output layer, so that the number can be smaller than the number of outputs.
I just do not understand how fewer number of neurons can generate a larger number of output values.
Regards,
Greg
The number of inputs corresponds to the number of features in dataset where the number of outputs corresponds to the number of classes.
For example, we have a data set with a size of (1000, 9) and each individual prefer to travel in first class, second or third class. The architecture is 9 neurons in input layer, neurons in hidden is random while the neurons in output is 3. I hope that this response will be efficient.
- Badges
- Science topic