jensandClaude Sonnet 5
2f56608c54
--num was declared as "Number of games to play" but the play() call used a hardcoded 10. Also give --num an int type so it works when passed on the CLI (range() would otherwise crash on the raw string), and make --verbose a store_true flag. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01PQS7ZHqxQfy5DhKcz2gHau
109 lines
3.1 KiB
Python
109 lines
3.1 KiB
Python
import numpy as np
|
|
from state import create_empty_state, to_state_string
|
|
from machine_player import MachinePlayer
|
|
from human_player import HumanPlayer
|
|
from player_provider import PlayerProvider
|
|
import json
|
|
import argparse
|
|
|
|
float_formatter = "{:.3f}".format
|
|
np.set_printoptions(formatter={'float_kind': float_formatter})
|
|
|
|
REPORT_INTERVAL = 1000
|
|
report_counter = REPORT_INTERVAL
|
|
|
|
|
|
def play(player_provider: PlayerProvider, k_max=10000, with_debug=False):
|
|
player_provider.set_debug(with_debug)
|
|
for k in range(0, k_max):
|
|
state = create_empty_state()
|
|
players = player_provider.choose(state)
|
|
move = 1
|
|
run = True
|
|
other_player = players[-1]
|
|
while run:
|
|
do_stop = False
|
|
this_reward = 0
|
|
other_reward = 0
|
|
for this_player in players:
|
|
if with_debug:
|
|
print(f"---------------------------------------------------")
|
|
print(f"- Game {k:06d}, Move {move} -----------------------------")
|
|
print(f"---------------------------------------------------")
|
|
last_state = state
|
|
state, has_moved = this_player.move(state)
|
|
if with_debug:
|
|
print(to_state_string(last_state, state))
|
|
if not has_moved:
|
|
this_reward = 0
|
|
other_reward = 0
|
|
do_stop = True
|
|
if with_debug:
|
|
print(f"{this_player.mark}: No more moves")
|
|
if this_player.has_won(state):
|
|
this_reward = 1
|
|
other_reward = 0
|
|
do_stop = True
|
|
if with_debug:
|
|
print(f"{this_player.mark}: Has won the game")
|
|
|
|
if do_stop:
|
|
this_player.end_game(state, this_reward)
|
|
other_player.end_game(state, other_reward)
|
|
run = False
|
|
break
|
|
|
|
other_player = this_player
|
|
move += 1
|
|
|
|
|
|
if __name__ == '__main__':
|
|
parser = argparse.ArgumentParser()
|
|
parser.add_argument("--verbose", default=False, action='store_true', help="Be verbose")
|
|
parser.add_argument("--train", default=False, help="Train before play")
|
|
parser.add_argument("--num", default=5, type=int, help="Number of games to play")
|
|
parser.add_argument("--human", default=False, help="X-player is human")
|
|
parser.add_argument("--load", default=True, help="Load experience from previous trainings")
|
|
args = parser.parse_args()
|
|
|
|
if args.load:
|
|
try:
|
|
with open("values_x.json", "r") as fp:
|
|
x_values = json.load(fp)
|
|
|
|
with open("values_o.json", "r") as fp:
|
|
o_values = json.load(fp)
|
|
except Exception:
|
|
x_values = {}
|
|
o_values = {}
|
|
else:
|
|
x_values = {}
|
|
o_values = {}
|
|
|
|
p_mx = MachinePlayer(mark='X', params=MachinePlayer.Params(p_explore=0.1, alpha=0.1), values=x_values)
|
|
p_mo = MachinePlayer(mark='O', params=MachinePlayer.Params(p_explore=0.1, alpha=0.1), values=o_values)
|
|
|
|
N = args.num
|
|
if args.train:
|
|
for n in range(0, N):
|
|
play(PlayerProvider(p_mx, p_mo), 1000, False)
|
|
|
|
# Values after training
|
|
# Convert and write JSON object to file
|
|
with open("values_x.json", "w") as fp:
|
|
json.dump(p_mx.values, fp, indent=0)
|
|
|
|
with open("values_o.json", "w") as fp:
|
|
json.dump(p_mo.values, fp, indent=0)
|
|
|
|
print(f"Iteration: {n*N:8d}")
|
|
|
|
if args.human:
|
|
p_hx = HumanPlayer(mark='X', name="Jens")
|
|
pp = PlayerProvider(p_hx, p_mo)
|
|
else:
|
|
pp = PlayerProvider(p_mx, p_mo)
|
|
|
|
play(pp, N, args.verbose)
|
|
|