How to customize number of hidden layers and number of Neurons in Neural Nets using Rapidminer?

vjmeena

Is there a way to customize number of hidden layers and number of Neurons in Neural Nets using Rapidminer?

varunm1

hi @vjmeena,

You can add hidden layers in an edit list of neural net operator parameter window. Give the number of neurons you need in hidden layers sizes and name layers 1, 2, .. depends on the number of layers you need. Adding layers is done by clicking "Add Entry" in the below image. If you want a single deep learning operator with different activation functions you can select deep learning operator and do the same there and you can choose various activations as well.

[Deleted User]

@varunm1
hi
the result of your suggestion is good for deep learning but for neural network it is not that much ok. also neural network has so many problems itself and i have to delete most of the row of data

varunm1

Hello @mbs

Sorry, I didn't get your point. Can you please explain in detail? What is the issue you are facing? 

[Deleted User]

@varunm1
according to your solution in this post I try to make less hidden layers with deep learning and also i did it for neural network but neural network most of the time show that my data has missing value in different parts but really my data is clean and it doesnt have any missing part so even with changing the setting of that i have to remove some parts of my data and even in this situation accuracy is around 50%
regards
mbs

varunm1

The solution here just refers to customizing hidden layers with neurons in a neural network operator and nothing related to the performance of networks. 

The deep learning operator is more customizable with different activation functions (that can be chosen) etc. They are much more complex compared to a simple neural net which is the reason you might get better results. Coming to the issue related to missing values, it's difficult to understand unless I see the data and your XML. Also, deep learning operator handles missing values (you can see in parameters) by either imputing or skipping them which might be the reason you are not getting an error.

Thanks

[Deleted User]

@varunm1
thank you for your recommendation