how many datasets are used to training neural network in one iteration?

If neural network has two inputs and one output, and there are 100 datasets for two inputs and one output, how many datasets are used in one iteration training. I use default setting for training set (70%), validation set (15%), and test set (15%). Does one iteration training uses 70% randomly selection training data? Who can give the answer? Thank you.

Answer 1

[Greg Heath]

 1 dataset

 100 examples

Each epoch the batch training default, trainlm, uses 70 training examples to update weights, output and training error; 15 examples to update validation output and validation error and the remaining 15 examples to update the test output and test error.

Hope this helps.

Thank you for formally accepting my answer

Greg

Answer 2

[Amend]

Thank you for your answer. For above question,if I use default setting (net.trainParam.epochs=1000), each time network randomly selects 70 samples in one epoch or one iteration to perform training until the error goal is met or the maximum number of epochs is reached, or overfitting is found, am I right?

[Greg Heath]

No. Once randomly chosen, the trn/val/tst sets remain fixed.

Answer 3

[Amend]

"Once randomly chosen, the trn/val/tst sets remain fixed.", do you mean:

1. trn/val/tst data sets keep fixed until the error goal is met, or the maximum number of epochs is reached, or overfitting is found.

or

2. trn/val/tst data sets keep fixed during one epoch or iteration. if error goal is not met, or not max epoch number, or not overfit, take another randomly choose to make another training. And repeat these steps until the error goal is met, or the maximum number of epochs is reached, or overfitting is found.

Which one?