Is it necessary to test a neural networks with data set unseen by ANN ??

1 visualización (últimos 30 días)
omar belhaj
omar belhaj el 26 de Feb. de 2015
Comentada: omar belhaj el 28 de Feb. de 2015
Purpose: A neural networks, when properly trianed must have ability to generalize and not behave just like a memory.
So, for me i used MLP( 7 inputs,1 ouputs, 1 hidden layer) type's of ANN with 13 years data set, i divided these data as following :
- 10 years data set to train my ANN
- 3 years unseen by ANN to test my model
My question is,can i delete this command 'net.divideparam.testRatio' since i used data unseen by ANN ????
net.divideParam.trainRatio = 75/100;
net.divideParam.valRatio = 15/100;
|net.divideParam.testRatio = 10/100; (I think this command is not necessary!!! can i delete it ??)

Iniciar sesión para responder a esta pregunta.

Respuesta aceptada

Greg Heath
Greg Heath el 27 de Feb. de 2015
Just put all of the data into the net.
Then choose the datadivision mode and indexing that suits the problem.
Nothing more is needed.
  1 comentario
omar belhaj
omar belhaj el 28 de Feb. de 2015
So, that I "ve understood from your answer for exemple for this case data set 13 years
net.divideParam.trainRatio = 75/100;( about 10 years are used for training)
net.divideParam.valRatio = 15/100; (about 2 years are used for validaition to avoid "overfitting")
|net.divideParam.testRatio = 10/100 ( about 1 year used to test our model, we can consired that unseen by NN !!)

Iniciar sesión para comentar.

Más respuestas (0)

Categorías

Más información sobre Deep Learning Toolbox en Help Center y File Exchange.

Etiquetas

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by