def turn(agent: Agent,
agent_policy_step: str,
foe: Foe,
rand="random") -> (str, str):
# Not currently very extensible. Oh well.
agent_action = agent.act(agent_policy_step)
new_states = (agent_action.resolve_action(foe) if rand == "random" else
agent_action.action_expectation(foe))
if agent_action.target_id == "self":
agent.update_states(new_states["target"])
else:
foe.update_states(new_states["target"])
foe_action = foe.act(rand)
new_states = (foe_action.resolve_action(agent) if rand == "random" else
foe_action.action_expectation(agent))
if foe_action.target_id == "self":
foe.update_states(new_states["target"])
else:
agent.update_states(new_states["target"])
foe.decrement_cooldowns()
foe_reaction = foe.react()
return (agent_action, foe_reaction)
def __init__(self, screen_dim):
self.screen_dim = screen_dim
self.background = [
StarsLayer(screen_dim, (100, 100, 105), 1, 3),
StarsLayer(screen_dim, (80, 80, 120), .5, 2),
StarsLayer(screen_dim, (55, 55, 55), .25, 1)
]
self.foreground = [
StarsLayer(screen_dim, (255, 255, 255), 1.5, 4, 22),
StarsLayer(screen_dim, (255, 255, 255), 6, 8, 7)
]
self.enemies = [Foe(screen_dim), Foe(screen_dim)]
self.exploding = []
self.bubbles = [bubble.BubbleFactory().create_random(screen_dim)]
self.food = []
def update_foe_belief(faux_foe: Foe, foe_reaction: str):
if "RAW" in foe_reaction:
print(foe_reaction)
faux_foe.hp = (14 - (len(foe_reaction) - 3)) * 10
return faux_foe
def spawn_foe(self):
randy = random.random()
if randy > SPAWN_RATE:
return
foe = Foe(self)
foe.set_color(0, 255, 0)
randy = random.uniform(0, 1.25)
foe.set_radius(randy * self.player.radius)
foe.choose_location_foe(self.screen_size, self.player)
self.foes_list.append(foe)
def __init__(
self,
agent: Agent = Agent(),
foe: Foe = Foe(),
reward_for_kill: float = 1000,
penalty_for_dying: float = -1000,
agent_hp_bonus: float = 0,
foe_hp_bonus: float = -0.5,
):
# See numpy.interp
self.agent_xp = np.arange(start=1, stop=agent.max_hp + 1)
self.reward_per_agent_hp = agent_hp_bonus * self.agent_xp
self.penalty_for_dying = penalty_for_dying
self.foe_xp = np.arange(start=1, stop=foe.max_hp + 1)
self.reward_per_foe_hp = foe_hp_bonus * self.foe_xp
self.reward_for_kill = reward_for_kill
def encounter(
agent=Agent,
foe=Foe,
max_turns=int,
forward_search_and_reward_kwargs={},
) -> pd.DataFrame:
"""
TODO: document me!!
:param agent:
:param foe:
:param max_turns:
:param forward_search_and_reward_kwargs:
- forward_search_kwargs:
- depth: int = 3,
- reward_kwargs:
- reward_for_kill: float = 1000,
- penalty_for_dying: float = -1000,
- agent_hp_bonus: float = 2,
- foe_hp_bonus: float = -1
:return:
"""
# Handle kwargs
forward_search_kwargs, reward_kwargs = __get_kwargs(
forward_search_and_reward_kwargs)
reward = Reward(agent, foe, **reward_kwargs)
utility = reward.get_reward(agent, foe)
faux_foe = Foe() # The belief state of our foe
# Arrays to hold encounter_stats
agent_policies = []
agent_spell_slots = []
agent_shields = []
agent_healths = []
foe_healths = []
foe_reactions = []
faux_foe_healths = []
forward_search_utilities = []
rewards = []
for i in range(max_turns):
agent_policy, forward_search_utility = forward_search(
agent=copy.deepcopy(agent),
foe=copy.deepcopy(faux_foe),
reward=reward,
utility=utility,
**forward_search_kwargs)
agent_action, foe_reaction = turn(agent, agent_policy, foe)
faux_foe = update_foe_belief(faux_foe, foe_reaction)
utility += reward.get_reward(agent, foe)
# Collect turn data into encounter_stats
agent_policies.append(agent_policy)
agent_spell_slots.append(agent.states["spell slots"])
agent_shields.append(agent.states["shield"])
agent_healths.append(agent.hp)
foe_healths.append(foe.hp)
foe_reactions.append(foe_reaction)
faux_foe_healths.append(faux_foe.hp)
forward_search_utilities.append(forward_search_utility)
rewards.append(utility)
if agent.hp <= 0 or foe.hp <= 0:
# end encounter if either dies
break
encounter_stats = pd.DataFrame({
"agent actions": agent_policies,
"agent spell slots": agent_spell_slots,
"agent shield": agent_shields,
"agent health": agent_healths,
"foe health": foe_healths,
"foe reactions": foe_reactions,
"faux foe health": faux_foe_healths,
"forward search utility": forward_search_utilities,
"utility": rewards,
})
return agent, foe, encounter_stats
import pandas as pd
from agent import Agent
from encounter import encounter
from foe import Foe
bad_guy = Foe()
good_guy = Agent()
# good_guy, bad_guy, encounter_stats = encounter(good_guy, bad_guy, 100)
good_guy, bad_guy, encounter_stats = encounter(
good_guy,
bad_guy,
100,
forward_search_and_reward_kwargs={
"forward_search": {
"depth": 25
},
"reward": {
"penalty_for_dying": -5000
},
})
## Note: Valerie moved plotting to test.ipynb here because reasons
def inc_enemies(self):
self.enemies.append(Foe(self.screen_dim))