def generate_data_rl(self, ngames=50, fname=''):
"""
generate a new batch of data with the latest prediction model
self.model_predict
rl vs. smithy bot
"""
vf = lambda x: self.model_predict.predict(x)
p1 = RLPlayer(vf)
p1.epsilon = self.epsilon
p1.record_history = 1
p2 = RLPlayer(vf)
p2.epsilon = self.epsilon
p2.record_history = 1
d_this = record_game(ngames, [p1, p2], fname)
self.add_data(d_this)
return d_this
def generate_data_rl(self, ngames=50, fname=''):
"""
generate a new batch of data with the latest prediction model self.model_predict
by playing against self.
"""
vf = lambda x: self.model.predict(x)
p2 = RLPlayer(vf)
p2.epsilon = self.epsilon
p2.record_history = 1
p2.include_action = 1
return self.generate_data_bot(p2, ngames, fname)
def generate_data_rl(self, ngames=50, fname=''):
"""
generate a new batch of data with the latest prediction model
self.model_predict
rl vs. smithy bot
"""
vbuy = lambda x: self.model_predict.predict(x)
#vact = lambda x: self.model_act.predict(x)
# p1 = BuyActRLplayer(vbuy, vact)
p1 = RLPlayer(vbuy)
p1.epsilon = self.epsilon
p1.record_history = 1
p1.include_action = 1
p2 = RLPlayer(vbuy)
p2.epsilon = self.epsilon
p2.record_history = 1
p2.include_action = 1
d_this, _ = self.record_game(ngames, [p1, p2], fname, verbose=1)
self.add_data(d_this)
return d_this
def compare_rl_bots(fn1, fn2, num_games=50, order=0):
# set up the two rl bots, and make them fight.
print('setting up bot1...')
dql = SarsaAgent()
dql.create_model_5layers()
data = dql.load_game_data('1game')
dql.fit(data)
dql.load_model(fn1)
p1 = RLPlayer(lambda x: dql.model.predict(x))
p1.name = fn1
p1.epsilon = 0
print('setting up bot2...')
dql2 = SarsaAgent()
dql2.create_model_5layers()
data = dql2.load_game_data('1game')
dql2.fit(data)
dql2.load_model(fn1)
p2 = RLPlayer(lambda x: dql2.model.predict(x))
p2.name = fn2
p2.epsilon = 0
print('fight!')
return compare_bots([p1, p2], num_games, order)
def generate_data_bot(self, bot, ngames=50, fname=''):
"""
generate a new batch of data with the latest prediction model self.model
rl vs. specified bot
"""
vf = lambda x: self.model.predict(x)
p1 = RLPlayer(vf)
p1.epsilon = self.epsilon
p1.record_history = 1
p1.include_action = 1
bot.record_history = 0
d_this = self.record_game(ngames, [p1, bot], fname)
self.add_data(d_this)
return d_this
def generate_data_smithy(self, ngames=50, fname=''):
"""
generate a new batch of data with the latest prediction model
self.model_predict
rl vs. smithy bot
"""
vbuy = lambda x: self.model_predict.predict(x)
# vact = lambda x: self.model_act.predict(x)
# p1 = BuyActRLplayer(vbuy, vact)
p1 = RLPlayer(vbuy)
p1.epsilon = self.epsilon
p1.record_history = 1
p1.include_action = 1
p2 = SmithyBot()
# try including smithy bot's data in the training.
p2.record_history = 0
d_this, _ = self.record_game(ngames, [p1, p2], fname)
self.add_data(d_this)
return d_this