Closed Skeen closed 7 years ago
You can use your own cost function in trainer, but it should by sync - no async requests. See more in docs: https://github.com/cazala/synaptic/wiki/Trainer predefined cost functions can be viewed here, as an example: https://github.com/cazala/synaptic/blob/a5fbefea5a69700384e2fe4c9fd82b2c50146c5f/src/trainer.js#L654-L676
Thanks, what about randomized trainingsets?
Can you please explain meaning of it? You mean shuffled set?
Hello,
Is it possible to train using a 'fitness' function instead of a training set?
I.e. have the network receive random input, generate an output, and then be 'rewarded' for the accuracy of it's output? This is how I've usually been working within the field of evolutionary algorithms, but I'm not sure how it maps directly to utilizing neural networks.