i used #wolframalpha to find weights and biases for #neuralnetwork to produce given output for some inputs.
my conditions were:
- for input 99, output should be 2
- for input 50, output should be 100
here is the query i gave to wolfram:
max(0, (a * max(0, (b * 99) + c) + d * max(0, (0.4 * 99) + 0.2)) + 0.3) == 2 and max(0, (a * max(0, (b * 50) + c) + d * max(0, (0.4 * 50) + 0.2)) + 0.3) == 100
task for wolfram was to find correct values for weights a, b and biases c, d
wolfram alpha was able to solve this equation and find suitable values:
a = 1
b = -2007/(995*a)+1
c = (5*a*b + 2007)/(10*a)
d = -5/2*(a*b + 2)