Added snakenet.rb, the tool used to train the AI for snake effect.
This commit is contained in:
parent
a3caaa1fef
commit
bd0041619a
344
src/tools/snakenet/snakenet.rb
Normal file
344
src/tools/snakenet/snakenet.rb
Normal file
@ -0,0 +1,344 @@
|
||||
#!/usr/bin/env ruby
|
||||
require 'rubygems'
|
||||
require 'pp'
|
||||
require 'thread/pool'
|
||||
|
||||
SEEDS = [
|
||||
#[0, 1, -1, 0, -1, 0, 1, -1, 1, -1, 1, 1, 1, 0, -1, 0, 0, 1, 0, -1, 0, -1, 1, 0, 0, 0, 0, 0, 0, 1, -1, 0, -1, -1, 0, 1, -1, 0, 0, 0, -1, -1, -1, 0, -1, 1, 0, -1, -1, 0, -1, 1, 0, 1, 1, 0, -1, -1, 0, -1, 1, 0, -1, 0],
|
||||
#[-1, 1, -1, 0, -1, 1, -1, -1, 1, 1, -1, 1, 1, -1, 1, 0, 0, -1, 0, 1, 0, -1, -1, 1, 0, 0, 0, -1, 0, 0, 1, -1, -1, -1, -1, -1, -1, 0, 0, -1, -1, 0, -1, 0, -1, 1, 1, 1, -1, 0, -1, 1, 0, 0, 0, -1, -1, -1, 0, -1, 1, 1, -1, 1],
|
||||
#[1, 1, -1, 0, 1, 1, 0, -1, 1, 1, 1, 1, 1, -1, -1, -1, 0, -1, 0, 1, 0, -1, 0, 1, 0, 0, 0, -1, 0, 1, 1, 1, -1, -1, -1, 1, -1, -1, 0, -1, -1, -1, -1, 0, -1, -1, 1, 1, 0, 0, -1, 1, 0, 0, 0, 1, -1, -1, 0, 0, 1, 1, 0, 1],
|
||||
[1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,1,0,0,0,1,-1,-1,-1,-1,-1,-1,-1,-1,-1,1,-1,1,-1,-1,-1,-1,0,0,0,0,0,0,-1,-1,0,0,0,1,0,0,-1,-1]
|
||||
|
||||
]
|
||||
|
||||
GAMES_PER_ROUND = 50
|
||||
|
||||
class Game
|
||||
WIDTH = 16
|
||||
HEIGHT = 10
|
||||
|
||||
POINTS_APPLE = 10
|
||||
POINTS_MOVING_CLOSER = 1
|
||||
POINTS_MOVING_FAR = -1.5
|
||||
|
||||
attr_reader :points, :dead, :ai, :length
|
||||
|
||||
def initialize(a)
|
||||
@ai = a
|
||||
@data = [0]*(WIDTH*HEIGHT)
|
||||
@dir = 0
|
||||
@pos = [WIDTH/2, HEIGHT/2]
|
||||
@data[(@pos[1] )*WIDTH + @pos[0]]=1
|
||||
@data[(@pos[1]+1)*WIDTH + @pos[0]]=2
|
||||
@data[(@pos[1]+2)*WIDTH + @pos[0]]=3
|
||||
@data[(@pos[1]+3)*WIDTH + @pos[0]]=4
|
||||
@length = 4
|
||||
@points = 0.0
|
||||
@dead = false
|
||||
@round = 0
|
||||
@last_apple_at = 0
|
||||
place_apple()
|
||||
end
|
||||
|
||||
def place_apple
|
||||
x=-1
|
||||
while @data[x]!=0 || x==-1
|
||||
x = rand(WIDTH*HEIGHT)
|
||||
end
|
||||
@apple = [x%WIDTH, x/WIDTH]
|
||||
@old_distance = apple_distance()
|
||||
end
|
||||
|
||||
def apple_distance
|
||||
return (@pos[0] - @apple[0]).abs + (@pos[1] - @apple[1]).abs
|
||||
end
|
||||
|
||||
def to_s
|
||||
str = @data.join("").gsub("0", " ")
|
||||
str[@apple[1]*WIDTH+@apple[0]] = "*"
|
||||
s = "+" + "-"*(WIDTH-@points.to_s.length-1)+" "+@points.to_s+"+\n"
|
||||
(0...HEIGHT).each do |y|
|
||||
s += "|" + str[y*WIDTH, WIDTH] + "|\n"
|
||||
end
|
||||
s += "+" + "-"*WIDTH+"+\n"
|
||||
return s
|
||||
end
|
||||
|
||||
def draw; puts to_s; puts; end
|
||||
|
||||
def loop
|
||||
return if @dead
|
||||
decision = @ai.decide(free?(@dir-1), free?(@dir), free?(@dir+1), apple?(@dir-1), apple?(@dir), apple?(@dir+1))
|
||||
@dir = (@dir + decision) % 4
|
||||
if (!free?(@dir))
|
||||
die()
|
||||
return
|
||||
end
|
||||
|
||||
move
|
||||
end
|
||||
|
||||
def ranking; @length*10 - (@dead ? 200 : 0) - (since_last_apple >= 160 ? 100 : 0); end
|
||||
|
||||
def move
|
||||
newpos = calc_new_pos(@pos, @dir)
|
||||
if newpos==@apple
|
||||
@length+=1
|
||||
@points += POINTS_APPLE
|
||||
@last_apple_at = @round
|
||||
place_apple
|
||||
end
|
||||
@data.each_with_index do |value, key|
|
||||
@data[key]=value+1 if value>0
|
||||
@data[key]=0 if value>=@length
|
||||
end
|
||||
@data[newpos[1]*WIDTH + newpos[0]] = 1
|
||||
@pos = newpos
|
||||
ad_d = apple_distance - @old_distance
|
||||
@old_distance = apple_distance()
|
||||
if (ad_d < 0)
|
||||
@points += POINTS_MOVING_CLOSER
|
||||
elsif (ad_d > 0)
|
||||
@points += POINTS_MOVING_FAR
|
||||
end
|
||||
@round+=1
|
||||
end
|
||||
|
||||
def since_last_apple; @round - @last_apple_at; end
|
||||
|
||||
def calc_new_pos(p, d)
|
||||
d = d%4
|
||||
np = p.dup
|
||||
case d
|
||||
when 0 then np[1]-=1
|
||||
when 1 then np[0]+=1
|
||||
when 2 then np[1]+=1
|
||||
when 3 then np[0]-=1
|
||||
end
|
||||
return np
|
||||
end
|
||||
|
||||
def free?(dir)
|
||||
newpos = calc_new_pos(@pos, dir)
|
||||
return newpos[0]>=0 && newpos[0]<WIDTH && newpos[1]>=0 && newpos[1]<HEIGHT && @data[newpos[1]*WIDTH + newpos[0]]==0
|
||||
end
|
||||
|
||||
def apple?(dir)
|
||||
dir = dir%4
|
||||
d_x = @apple[0] - @pos[0]
|
||||
d_y = @apple[1] - @pos[1]
|
||||
case dir
|
||||
when 0 then return @apple[1]<@pos[1]
|
||||
when 1 then return @apple[0]>@pos[0]
|
||||
when 2 then return @apple[1]>@pos[1]
|
||||
when 3 then return @apple[0]<@pos[0]
|
||||
#when 0 then return d_y<0 && d_x.abs<d_y.abs
|
||||
#when 1 then return d_x>0 && d_x.abs>d_y.abs
|
||||
#when 2 then return d_y>0 && d_x.abs<d_y.abs
|
||||
#when 3 then return d_x<0 && d_x.abs>d_y.abs
|
||||
end
|
||||
end
|
||||
|
||||
def die
|
||||
@dead = true
|
||||
end
|
||||
|
||||
def stopped?; since_last_apple >= WIDTH*HEIGHT*1.5; end
|
||||
|
||||
def ai_ranking; ai.ranking; end
|
||||
end
|
||||
|
||||
class AI
|
||||
attr_reader :weights
|
||||
attr_accessor :ranking, :rounds, :count_dead, :count_stopped, :sum_length
|
||||
|
||||
def initialize(w=nil)
|
||||
reset()
|
||||
@rounds = 1
|
||||
if w==nil
|
||||
@weights = Array.new(2**6){ rand(3)-1 }
|
||||
else
|
||||
@weights = w
|
||||
end
|
||||
end
|
||||
|
||||
def reset
|
||||
@ranking = 0.0
|
||||
@count_dead = 0
|
||||
@count_stopped = 0
|
||||
@sum_length = 0
|
||||
end
|
||||
|
||||
def add_ranking(g)
|
||||
@ranking += g.ranking
|
||||
@count_dead += 1 if g.dead
|
||||
@count_stopped += 1 if g.stopped?
|
||||
@sum_length += g.length
|
||||
end
|
||||
|
||||
def decide(left_free, straight_free, right_free, apple_left, apple_straight, apple_right)
|
||||
input = apple_right ? 1 : 0 | (apple_straight ? 1<<1 : 0) | (apple_left ? 1<<2 : 0) | (right_free ? 1<<3 : 0) | (straight_free ? 1<<4 : 0) | (left_free ? 1<<5 : 0)
|
||||
w = @weights[input]
|
||||
return w
|
||||
=begin
|
||||
inputs = [left_free, straight_free, right_free, apple_left, apple_straight, apple_right]
|
||||
outputs = [0, 0, 0]
|
||||
(0...18).each do |x|
|
||||
o = x/6
|
||||
i = x%6
|
||||
outputs[o] += (inputs[i] ? 1.0 : 0.0) * @weights[x]
|
||||
end
|
||||
max = 0
|
||||
take = 0
|
||||
(0...3).each do |x|
|
||||
if outputs[x]>max
|
||||
max = outputs[x]
|
||||
take = x
|
||||
end
|
||||
end
|
||||
return take-1
|
||||
=end
|
||||
end
|
||||
|
||||
def evolve
|
||||
w = @weights.dup
|
||||
loop do
|
||||
if rand(2)==0
|
||||
# swap
|
||||
i1 = rand(64)
|
||||
i2 = rand(64)
|
||||
next if w[i1]==w[i2]
|
||||
temp = w[i1]
|
||||
w[i1] = w[i2]
|
||||
w[i2] = temp
|
||||
break
|
||||
else
|
||||
# set new
|
||||
i = rand(64)
|
||||
v = rand(3)-1
|
||||
next if w[i]==v
|
||||
w[i] = v
|
||||
break
|
||||
end
|
||||
end
|
||||
|
||||
# Sanity checks
|
||||
(0...(2**6)).each do |i|
|
||||
w[i] = rand(3)-1 if (i&(1<<4)==0) && w[i]==0
|
||||
w[i] = rand(3)-1 if (i&(1<<3)==0) && w[i]==1
|
||||
w[i] = rand(3)-1 if (i&(1<<5)==0) && w[i]==-1
|
||||
end
|
||||
#i = rand(2**6)
|
||||
#v=0
|
||||
#loop do
|
||||
# v = rand(3)-1
|
||||
# break if v!=w[i]
|
||||
#end
|
||||
#w[i]=v
|
||||
|
||||
#x = rand(18)
|
||||
#w[x] += rand(0.4)-0.2
|
||||
#w[x] = 0.0 if w[x]<0
|
||||
#w[x] = 1.0 if w[x]>1
|
||||
return AI.new(w)
|
||||
end
|
||||
|
||||
def merge(ai)
|
||||
w = @weights.dup
|
||||
w2 = ai.weights
|
||||
(0...(2**6)).each do |i|
|
||||
if rand(5)==0
|
||||
w[i] = w2[i]
|
||||
end
|
||||
end
|
||||
return AI.new(w)
|
||||
end
|
||||
|
||||
def simplified
|
||||
#res = []
|
||||
#(0...(2**6)).each do |i|
|
||||
# args = i.to_s(2).rjust(6, '0').split("").map{|c| c=='1'}
|
||||
# res << decide(*args)
|
||||
#end
|
||||
v = 0
|
||||
count = 0
|
||||
result = []
|
||||
(0...(2**6)).each do |i|
|
||||
v = v<<2 | @weights[i]
|
||||
count += 1
|
||||
if count==16
|
||||
result << "0x" + v.to_s(16)
|
||||
count = 0
|
||||
v = 0
|
||||
end
|
||||
end
|
||||
return "uint32_t weights[4] = {" + result.join(", ") + "};"
|
||||
end
|
||||
|
||||
def dump
|
||||
puts "Data:"
|
||||
puts "int8_t _decisions[64] = {#{@weights.join(", ")}};"
|
||||
#puts "Simplified: #{simplified}"
|
||||
end
|
||||
end
|
||||
|
||||
ais = []
|
||||
round = 1
|
||||
games = []
|
||||
(0...50).each do |x|
|
||||
ais[x] = AI.new#(SEEDS.sample)
|
||||
end
|
||||
|
||||
best_old_ai = nil
|
||||
begin
|
||||
loop do
|
||||
GAMES_PER_ROUND.times do
|
||||
(0...50).each do |x|
|
||||
games[x] = Game.new(ais[x])
|
||||
end
|
||||
|
||||
#pool = Thread.pool(16)
|
||||
games.each do |g|
|
||||
|
||||
# pool.process do
|
||||
15_000.times do
|
||||
g.loop
|
||||
break if g.dead || g.stopped?
|
||||
end
|
||||
g.ai.add_ranking(g)
|
||||
|
||||
end
|
||||
#end
|
||||
#pool.shutdown
|
||||
end
|
||||
|
||||
games_sorted = games.sort_by(&:ai_ranking).reverse.take(5)
|
||||
g = games_sorted[0]
|
||||
|
||||
if true#round%10==0
|
||||
puts "Round %5d: %7.1f points, length %3.0f, %3.0f%% stopped, %3.0f%% dead" % [round, g.ai_ranking / GAMES_PER_ROUND, g.ai.sum_length.to_f / GAMES_PER_ROUND, g.ai.count_stopped.to_f / GAMES_PER_ROUND * 100, g.ai.count_dead.to_f / GAMES_PER_ROUND * 100]
|
||||
end
|
||||
|
||||
best_old_ai = g.ai
|
||||
|
||||
ais = []
|
||||
games_sorted.each do |g|
|
||||
g.ai.reset
|
||||
g.ai.rounds += 1
|
||||
ais << g.ai
|
||||
9.times do
|
||||
ais << g.ai.evolve
|
||||
end
|
||||
end
|
||||
|
||||
5.times do
|
||||
ais << games_sorted[0].ai.merge(games_sorted[1].ai)
|
||||
end
|
||||
5.times do
|
||||
ais << games_sorted[1].ai.merge(games_sorted[0].ai)
|
||||
end
|
||||
round+=1
|
||||
end
|
||||
rescue SystemExit, Interrupt
|
||||
best_old_ai.dump
|
||||
end
|
||||
Loading…
Reference in New Issue
Block a user