0
$\begingroup$

I'm using Orange Data Mining in a regression analysis applying Artificial Neural Network (ANN). Some works suggest defining the number of neurons in the input layer as the number of variables. The Orange ANN uses sklearn’s Multi-layer Perceptron algorithm (MLP); however, I can only set the number of neurons per hidden layer (e.g., 4, 3 and 2; tree layers with 4, 3 and 2 neurons). Should I assume a hidden layer as the input layer and set the number of neurons as variable numbers? Does the MLP set the input layer automatically?

$\endgroup$

2 Answers 2

1
$\begingroup$

I am not familiar with the software you are using, but the number of neurons on the input layer depends on the number of features in the data. What follows, the software can just count the columns and use them to set the number of neurons on the input layer. The hidden layer is not an input layer, it is unlikely for them to name things in a way that confuses those two types of layers.

$\endgroup$
3
  • $\begingroup$ thanks of your answer. Maybe I wasn’t clear enough. So let me try to clarify. Orange name the parameters correctly, you can check this in this orange3.readthedocs.io/projects/orange-visual-programming/en/…. My doubt is if I can use the hidden layer as the input one due to the absence of a specific parameter in sklearn’s Multi-layer Perceptron algorithm (MLP), which is used in Orange. As you commented, I understand that hidden layer is not an input layer. Thus, I should not set a hidden layer as the number of features :) $\endgroup$
    – kvratto
    Commented Jun 21, 2021 at 13:07
  • $\begingroup$ @kvratto How exactly would you use it as an input layer? It will set up the input layer by itself unless I am missing something about the software itself. $\endgroup$
    – Tim
    Commented Jun 21, 2021 at 13:08
  • $\begingroup$ I didn't find if the sklearn’s Multi-layer Perceptron algorithm (and consequently Orange) set up the input layer by itself. I thought the algorithm just skipped this step (set up the input layer) or ran with a number of neurons in the standard input layer. I truly appreciate your comments. $\endgroup$
    – kvratto
    Commented Jun 21, 2021 at 13:20
0
$\begingroup$

If i want to share my failure in use of Orange then i have to put entire workflow here. In Orange u have to put hidden layer's if then more than one your self. By default it is set to "100,". If u want more layers then u put values like that "100, 120, 120, 100," in corresponding tab in widget. This means u have 4 hidden layers with nodes according to specified numbers provided. The input and output levels are not adjustable. There is upside to this options, one can play with different NN architecture to see which fits better. Downside if u don't find first good up settings to fit model u can loose your mind trying. There is no AutoFit for Learner mode in Orange for NN to guide you to proper way. And here im failing to find corresponding setting for NN for my dataset. If u aren't developer u most likely would fail in use of NN. Also problems with numbers dataset has only whole numbers but NN's throwing out decimal ones. Also there is no limitation to numbers for throwing out numbers bigger that max number provided in dataset. I get better results more than 80% accuracy with the same dataset in few years old NeoNeuro data minig software. That's very weird.

$\endgroup$
1

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.