CODEWITHSUNDEEP
javamachine-learningneural-network
ahodder
ahodder

Reputation: 11439

How do you train a FeedForward Supervised Learning NeuralNet?

Suppose I am building a Neural Net to play tic-tac-toe. As I understand it, a good design for the net would be 9 input neurons [one for each square in the grid] 3 hidden layer neurons and 9 output neurons [one for each potential move]. Now that the Net is built, how do you read the output of the net?

As I understand it, to train the net, I would send the net the game board [via the inputs] everytime I need it to make a decision. But how do I obtain that decision?

Upvotes: 2

Views: 222

Answers (1)

jayunit100
jayunit100

Reputation: 17648

The key thing to realize here is that multiple output neural networks change their state to reflect a solution space, but they do not typically give you a hard and fast, final decision.

Ultimately, a neural net doesnt give you "just one answer" , but rather, it modifies its internal state to reflect a probabilistic landscape of solutions.

If you just want a single answer, then you will have to have a nueral net with only one output node.

There is no direct, right answer here

The question you have is actually quite sophisticated - the science of choosing an answer from a nueral net is an entire field of study, in and unto itself :

For some other insights, check out https://mathoverflow.net/questions/10697/methods-for-choosing-a-result-from-a-multiple-output-node-neural-network .\

Also, scan other resources for decision making methods for multiple-output neural nets.

Upvotes: 3

Related Questions