amitai5 / backprop-vs-genetics Goto Github PK

Some of the min and max values for the Activation Functions are not exactly accurate. Go through them all and make sure they are in the "range of meaning" for each one.

The XOR test from before while it still was technically valid, the starting position for the neural networks was not a normal distribution. This puts Backpropagation at a disadvantage making it harder for it to find a global minimum.

Fully implement the binary step activation function and its inverse.
Source: https://en.wikipedia.org/wiki/Activation_function

Fully implement the inverse square root activation function and its inverse.
Source: https://en.wikipedia.org/wiki/Activation_function

Activation Function Range:

Remove the parallel for loops in the computation of each neural network's value.

Instead parrallelize the Genetics and Backpropagation learning algorithms so that they calculate error/fitness Neural Network in parallel.

The Classification test from before while it still was technically valid, the starting position for the neural networks was not a normal distribution. This puts Backpropagation at a disadvantage making it harder for it to find a global minimum.

The code under the classification test is quite messy as I was just trying to get it to work. Take some time and clean up the code, make it more readable.

The Genetics algorithm seems to be getting stuck in multiple plateaus which could simply be a characteristic of the Genetics Algorithm itself, but still needs some fine tuning to make sure it is not something I am doing wrong.