https://github.com/gmashtalyar/AI-Game
Inputs
Bird Y, Top Pipe, Bottom Pipe
Connections
Inputs are connected to output with weight.
Outputs
Jump?
Activation funciton
TanH
Populations size
100 - random neural networks
Fitness Function
Distance
Max generations
30
Weighted sum = (x*weight) + (y*weight) +Bias
F = Activation Function
Output = F(weighted sum)
F = TanH
Weighted sum = -3500 -230 + 375 + 600
Output = F(-3500 - 230 + 375 + 600)
TanH(-2755) = -1
If output > 0.5 => jump else don’t jumphttps://github.com/gmashtalyar/AI-Game/blob/master/config-feedforward.txt
config
[NEAT]
fitness_critetion = > how we determine to choose best inputs (min/max/mean)
fitness_threshold = > how much fitness is enough
pop_size = > population size
reset_on_extinction = > do we reset values if all species are extinct
[DefaultGenome]
activation_default = > tanh
activation_mutate_rate => probably of new random activation number
activation_options = tanh
# node bias options
bias_min_value
bias_max_value
bias_init_mean
bias_init_stdev
bias_mutate_power
bias_mutate_rate
bias_replace_rate
# connection add/remove rates
conn_add_prob
conn_delete_prob
# connection enable options
enabled_default = > connections may be set to enable
enabled_mutate_rate
feed_forward
initial_connection
node_add_prob
node_delete_prob
# netword parameters
num_hidden
num_inputs
num_outputshttps://neat-python.readthedocs.io/en/latest/config_file.html#neat-section
.