To Do List:

• Layer Class

  • stores no. of incoming nodes, no. of outgoing nodes, weights, and biases
  • Calculate Output Function:
    • takes list of inputs (outputs of previous layer)
    • applies weight and bias associated with it
    • return list of values as output of layer (output of each node)

• Neural Network Class

  • creates and stores all the layers
  • Calculate Output Function:
    • initial input will be game parameters (dist from bars, goal)
    • passed to layer to calculate its output, which will act as input to next layer
    • output of final layer will the output of our NN

• Activation Function

  • apply ReLU or Sigmoid function on input

• Mutation Function

  • 90% chance for any parameter to mutate
  • start of generation should have more drastic mutations, while later on mutations should be gentler
  • maybe set an initial rate which decreases as the generations go on
  • rate of decrement could be fraction of generations over max generations

• Population Class

  • input is total population
  • those many dots and neural networks are initialised
  • Update Function
    • input = (distG, distB1, distB2)
    • input is passed to neural network of each dot
    • output = (moveup, moveleft, moveright)
    • dot moves any direction which has output > 1
    • no return value

• New Generation Function

  • saves 5 best dots from previous generation and uses them to create new population of 45 new dots while applying mutations
  • best 5 bots are added to this population
  • each parameter can come from any old dot probability of using old dots based on score