we got a job to give again. We need a neural network based software in PHP open source.
Now we would need a neural network (you can use whatever NN-lib you believe works the best, but if it needs installation you would need to install on our server or at least give a how-to-install) that is setup like that:
input layer: about 30 values
output layer: 1 value
Now how many perceptrons and stuff you use is up to you and your experience with neural networks.
We will give you about 4000 training value-pairs like e.g. Value1=1, Value2=300 and so on results in Output Value = 4.324
Now the target is that:
After the neural network is trained and setup it should give for at least 80% of the training parameters the correct results (even though there is a quite big aberation of 30+_ possible). So what I am meaning by that, if the real output value for a value-pair is e.g. 100, the neural network giving 70 or 130 would be ok as well.