def _reset(self):
self.env.reset()
self.model = Model(
self.env) # warning: this breaks if env resets again
start = self.Node(self.env._state, [], 0, False)
frontier = PriorityQueue(key=self.eval_node(
noisy=True)) # this is really part of the Meta Policy
frontier.push(start)
reward_to_state = defaultdict(lambda: -np.inf)
best_done = None
# Warning: state is mutable (and we mutate it!)
self._state = self.State(frontier, reward_to_state, best_done)
return self._state
from agents import Model, Agent, run
miner_model = Model("MinerBots", 100, 100)
miner_model.add_agent(Agent())
run(miner_model)
from random import randint
from agents import Model, Agent, AgentShape, run
miner_model = Model("MinerBots", 100, 100)
class Robot(Agent):
def setup(self, model):
self.color = (100, 100, 100)
self.direction = randint(0, 359)
self.loaded = False
self.x = model.width / 2
self.y = model.height / 2
def step(self, model):
if self.loaded:
self.point_towards(model.width / 2, model.height / 2)
else:
self.direction += randint(0, 20) - 10
self.forward()
self.speed = model.speed_factor
t = self.current_tile()
if t.info["has_mineral"] and not self.loaded:
t.info["has_mineral"] = False
t.color = (200, 100, 0)
self.color = (100, 100, 255)
self.loaded = True
class Homebase(Agent):
def setup(self, model):
class Astar(Policy):
"""A* search finds the shortest path to a goal."""
def __init__(self, heuristic):
assert 0 # this implementation is incorrect
super().__init__()
self.heuristic = heuristic
self.plan = iter(())
def start_episode(self, state):
self.history = Counter()
self.model = Model(self.env)
def act(self, state):
# return self.env.action_space.sample()
self.history[state] += 1
try:
return next(self.plan)
except StopIteration:
self.plan = iter(self.make_plan(state))
return next(self.plan)
def eval_node(self, node):
if not node.path:
return np.inf # the empty plan has infinite cost
obs = self.env._observe(node.state)
value = 0 if node.done else self.heuristic(self.env, obs)
boredom = -0.1 * self.history[obs]
score = node.reward + value + boredom
return -score
def make_plan(self, state, expansions=5000):
Node = namedtuple('Node', ('state', 'path', 'reward', 'done'))
eval_node = self.eval_node
start = Node(self.env._state, [], 0, False)
frontier = PriorityQueue(key=eval_node)
frontier.push(start)
reward_to_state = defaultdict(lambda: -np.inf)
# import IPython; IPython.embed()
best_finished = start
def expand(node):
# print(node.state, node.reward, self.rts[node.state], V(env._observe(node.state)))
# time.sleep(0.1)
nonlocal best_finished
# best_finished = min((best_finished, node), key=eval_node)
s0, p0, r0, _ = node
for a, s1, r, done in self.model.options(s0):
node1 = Node(s1, p0 + [a], r0 + r, done)
if node1.reward <= reward_to_state[s1]:
# print('abandon')
pass
continue # cannot be better than an existing node
# self.save('node', node)
reward_to_state[s1] = node1.reward
if done:
best_finished = min((best_finished, node1), key=eval_node)
else:
frontier.push(node1)
for i in range(expansions):
self.save('frontier', [n[1].state for n in frontier])
if frontier:
expand(frontier.pop())
else:
break
if frontier:
# plan = min(best_finished, frontier.pop(), key=eval_node)
plan = frontier.pop()
raise RuntimeError('No plan found.')
else:
plan = best_finished
# choices = concat([completed, map(get(1), take(100, frontier))])
# plan = min(choices, key=eval_node(noisy=True))
# self.log(
# i,
# len(plan.path),
# -round(eval_node(plan, noisy=False), 2),
# plan.done,
# )
# self._trace['paths'].append(plan.path)
self.save('plan', plan)
return plan.path
class ValSearchPolicy(Policy):
"""Searches for the maximum reward path using a model."""
def __init__(self,
V,
replan=False,
epsilon=0,
noise=1,
anneal=1,
**kwargs):
super().__init__(**kwargs)
self.V = V
self.replan = replan
self.epsilon = epsilon
self.noise = noise
self.anneal = anneal
self.history = None
self.model = None
self.plan = None
def start_episode(self, state):
self.history = Counter()
self.model = Model(self.env)
self.plan = iter(()) # start with no plan
def finish_episode(self, trace):
self.ep_trace['berries'] = self.env._observe()[-1]
def act(self, state):
# return self.env.action_space.sample()
self.history[state] += 1
try:
if self.replan:
raise StopIteration()
else:
return next(self.plan)
except StopIteration:
self.plan = iter(self.make_plan(state))
return next(self.plan)
def make_plan(self, state, expansions=2000):
Node = namedtuple('Node', ('state', 'path', 'reward', 'done'))
env = self.env
V = memoize(self.V.predict)
self.node_history = []
def eval_node(node, noisy=False):
if not node.path:
return np.inf # the empty plan has infinite cost
obs = env._observe(node.state)
noise = np.random.rand() * (
self.noise * self.anneal**self.i_episode) if noisy else 0
value = 0 if node.done else V(obs)[0]
boredom = -0.1 * self.history[obs]
score = node.reward + value + noise + boredom
return -score
start = Node(env._state, [], 0, False)
frontier = PriorityQueue(key=eval_node)
frontier.push(start)
reward_to_state = defaultdict(lambda: -np.inf)
reward_to_state[start.state] = 0
best_finished = start
def expand(node):
nonlocal best_finished
best_finished = min((best_finished, node), key=eval_node)
s0, p0, r0, _ = node
for a, s1, r, done in self.model.options(s0):
node1 = Node(s1, p0 + [a], r0 + r, done)
if node1.reward <= reward_to_state[s1]:
continue # cannot be better than an existing node
self.node_history.append({
'path':
node1.path,
'r':
node1.reward,
'b':
self.env._observe(node1.state)[-1],
'v':
-eval_node(node1)
})
reward_to_state[s1] = node1.reward
if done:
best_finished = min((best_finished, node1), key=eval_node)
else:
frontier.push(node1)
for i in range(expansions):
if frontier:
expand(frontier.pop())
else:
break
if frontier:
plan = min(best_finished, frontier.pop(), key=eval_node)
else:
plan = best_finished
# choices = concat([completed, map(get(1), take(100, frontier))])
# plan = min(choices, key=eval_node(noisy=True))
self.log(
i,
len(plan.path),
-round(eval_node(plan, noisy=False), 2),
plan.done,
)
# self._trace['paths'].append(plan.path)
return plan.path
for tile in model.tiles:
tile.color = (tile.info["value"], tile.info["value"], 0)
seekers = set([Seeker() for i in range(200)])
model.add_agents(seekers)
def step(model):
for tile in model.tiles:
if tile.info["value"] > 255:
tile.info["value"] = 0
tile.color = (tile.info["value"], tile.info["value"], 0)
for agent in model.agents:
agent.step(model)
def invisible(model):
model.invisible = not model.invisible
if model.invisible:
for a in model.agents:
a.size = 0
else:
for a in model.agents:
a.size = 4
urban_model = Model("urban", 50, 50)
urban_model.add_button("Setup", setup)
urban_model.add_toggle_button("Go", step)
urban_model.add_button("Invisible", invisible)
run(urban_model)
from agents import Model, Agent, run
epidemic_model = Model("Epidemimodel", 100, 100)
epidemic_model.add_agent(Agent())
run(epidemic_model)
model.Infectious -= 1
model.Recovered += 1
self.color = (0, 0, 200)
self.category = 2
def model_setup(model):
model.reset()
model.Susceptible = 0
model.Infectious = 0
model.Recovered = 0
for person in range(100):
model.add_agent(Person())
def model_step(model):
for person in model.agents:
person.step(model)
model.update_plots()
epidemic_model = Model("Epidemimodel", 100, 100)
epidemic_model.add_button("Setup", model_setup)
epidemic_model.add_button("Go", model_step, toggle=True)
epidemic_model.line_chart(
["Susceptible", "Infectious", "Recovered"], [(0, 200, 0), (200, 0, 0), (0, 0, 200)]
)
run(epidemic_model)
# Jump there
if new_tile is not None:
self.jump_to_tile(new_tile)
# Does nothing, if all tiles are occupied
def step(self, model):
self.move()
def setup(model):
model.reset()
model.initial_bugs = 100
# Create and add agents
for i in range(int(model.initial_bugs)):
model.add_agent(Bug())
def step(model):
# Move all agents
for agent in model.agents:
agent.step(model)
stupid_model = Model("Basic StupidModel (stupid01)", 100, 100, tile_size=5)
stupid_model.add_button("setup", setup)
stupid_model.add_button("step", step)
stupid_model.add_toggle_button("go", step)
run(stupid_model)
class MetaBestFirstSearchEnv(gym.Env):
"""A meta-MDP for best first search with a deterministic transition model."""
Node = namedtuple('Node', ('state', 'path', 'reward', 'done'))
State = namedtuple('State', ('frontier', 'reward_to_state', 'best_done'))
TERM = 'TERM'
def __init__(self, env, eval_node, expansion_cost=0.01):
super().__init__()
self.env = env
self.expansion_cost = -abs(expansion_cost)
# This guy interacts with the external environment, what a chump!
self.surface_agent = Agent()
self.surface_agent.register(self.env)
self.eval_node = eval_node
def _reset(self):
self.env.reset()
self.model = Model(
self.env) # warning: this breaks if env resets again
start = self.Node(self.env._state, [], 0, False)
frontier = PriorityQueue(key=self.eval_node(
noisy=True)) # this is really part of the Meta Policy
frontier.push(start)
reward_to_state = defaultdict(lambda: -np.inf)
best_done = None
# Warning: state is mutable (and we mutate it!)
self._state = self.State(frontier, reward_to_state, best_done)
return self._state
def _step(self, action):
"""Expand a node in the frontier."""
if action is self.TERM:
# The return of one episode in the external env is
# one reward in the MetaSearchEnv.
trace = self._execute_plan()
external_reward = trace['return']
return None, external_reward, True, {'trace': trace}
else:
return self._expand_node(action), self.expansion_cost, False, {}
def _execute_plan(self):
frontier, reward_to_state, best_done = self._state
if not best_done:
raise RuntimeError('Cannot make plan.')
policy = FixedPlanPolicy(best_done.path)
self.surface_agent.register(policy)
trace = self.surface_agent.run_episode(reset=False)
return trace
# elif frontier:
# plan = min(best_done, frontier.pop(), key=eval_node)
# plan = frontier.pop()
def _expand_node(self, node):
frontier, reward_to_state, best_done = self._state
s0, p0, r0, _ = node
for a, s1, r, done in self.model.options(s0):
node1 = self.Node(s1, p0 + [a], r0 + r, done)
if node1.reward <= reward_to_state[s1] - 0.002:
continue # cannot be better than an existing node
reward_to_state[s1] = node1.reward
if done:
best_done = max((best_done, node1),
key=self.eval_node(noisy=False))
else:
frontier.push(node1)
self._state = self.State(frontier, reward_to_state, best_done)
return self._state
from agents import Model, Agent, run
def model_setup(model):
model.reset()
model.add_agent(Agent())
epidemic_model = Model("Epidemimodel", 100, 100)
epidemic_model.add_button("Setup", model_setup)
run(epidemic_model)
model.remove_destroyed_agents()
# TODO: Stop after 1000 iterations
if model.agent_count() == 0:
model.pause()
def close(model):
if file_handle:
file_handle.close()
stupid_model = Model(
"StupidModel - reading habitat data from file (stupid15)",
100,
100,
tile_size=3,
cell_data_file="stupid.cell",
)
stupid_model.add_button("setup", setup)
stupid_model.add_button("step", step)
stupid_model.add_toggle_button("go", step)
stupid_model.add_controller_row()
stupid_model.add_slider("initial_bugs", 100, 10, 300)
stupid_model.add_controller_row()
stupid_model.add_slider("max_food_eat", 1.0, 0.1, 1.0)
stupid_model.add_controller_row()
stupid_model.add_slider("initialBugSizeMean", 1, 0, 10)
stupid_model.add_slider("initialBugSizeSD", 5, 0, 10)
stupid_model.histogram("grow_size", 0, 10, 5, (0, 0, 0))
stupid_model.line_chart(["current_bugs"], [(0, 0, 0)])
other.update_visual()
self.util = self.utility()
def setup(model):
model.reset()
model.clear_plots()
model.total_util = 0
model.movespeed = 0.2
people = set([Person() for i in range(20)])
model.add_agents(people)
def step(model):
model.total_util = 0
for a in model.agents:
a.step(model)
model.update_plots()
model.remove_destroyed_agents()
bnb_model = Model("Bread & butter economy", 50, 50)
bnb_model.add_button("Setup", setup)
bnb_model.add_button("Step", step)
bnb_model.add_button("Go", step, toggle=True)
bnb_model.add_controller_row()
bnb_model.add_slider("movespeed", 0.1, 0.1, 1)
bnb_model.line_chart(["total_util"], [(0, 0, 0)])
bnb_model.agent_line_chart("util")
run(bnb_model)
def step(self, model):
self.direction += randint(-10, 10)
self.forward()
if self.category == 1:
for agent in self.agents_nearby(12):
agent.infect(model)
def infect(self, model):
self.color = (200, 0, 0)
self.category = 1
def model_setup(model):
model.reset()
for person in range(100):
model.add_agent(Person())
def model_step(model):
for person in model.agents:
person.step(model)
epidemic_model = Model("Epidemimodel", 100, 100)
epidemic_model.add_button("Setup", model_setup)
epidemic_model.add_button("Go", model_step, toggle=True)
run(epidemic_model)
from random import randint
from agents import Model, Agent, run
miner_model = Model("MinerBots", 100, 100)
class Robot(Agent):
def setup(self, model):
self.color = (100, 100, 100)
self.direction = randint(0, 359)
def step(self, model):
self.direction += randint(0, 20) - 10
self.forward()
def setup(model):
model.reset()
for x in range(10):
model.add_agent(Robot())
def step(model):
for ag in model.agents:
ag.step(model)
miner_model.add_button("Setup", setup)
miner_model.add_toggle_button("Go", step)
def setup(model):
model.reset()
model.clear_plots()
model.total_util = 0
model.BNP = 0
people = set([Person() for i in range(20)])
model.add_agents(people)
def step(model):
model.BNP = 0
for a in model.agents:
a.step(model)
model.BNP += a.utility()
model.update_plots()
model.remove_destroyed_agents()
bnb_model = Model("Bread & butter economy during pandemic", 50, 50)
bnb_model.add_button("Setup", setup)
bnb_model.add_button("Step", step)
bnb_model.add_button("Go", step, toggle=True)
bnb_model.add_controller_row()
bnb_model.add_slider("movespeed", 0.5, 0.1, 1)
bnb_model.add_checkbox("Decay")
bnb_model.line_chart(["BNP"], [(0, 0, 0)])
bnb_model.show_direction = False
run(bnb_model)
from agents import Model, run
miner_model = Model("MinerBots", 100, 100)
run(miner_model)
from random import randint
from agents import Model, Agent, run
miner_model = Model("MinerBots", 100, 100)
class Robot(Agent):
def setup(self, model):
self.color = (100, 100, 100)
self.direction = randint(0, 359)
def step(self, model):
self.direction += randint(0, 20) - 10
self.forward()
self.speed = model.speed_factor
def setup(model):
model.reset()
for x in range(10):
model.add_agent(Robot())
model.speed_factor = 1
for t in model.tiles:
if randint(0, 50) == 50:
t.color = (0, 255, 255)
t.info["has_mineral"] = True
else:
t.color = (200, 100, 0)
t.info["has_mineral"] = False
from agents import Model, Agent, run
miner_model = Model("MinerBots", 100, 100)
def setup(model):
model.reset()
model.add_agent(Agent())
miner_model.add_button("Setup", setup)
run(miner_model)
model.update_plots()
model.remove_destroyed_agents()
# TODO: Stop after 1000 iterations
if model.agent_count() == 0:
model.pause()
def close(model):
if file_handle:
file_handle.close()
stupid_model = Model(
"StupidModel w. gauss distribution of sizes (stupid14)",
100,
100,
tile_size=5,
)
stupid_model.add_button("setup", setup)
stupid_model.add_button("step", step)
stupid_model.add_toggle_button("go", step)
stupid_model.add_controller_row()
stupid_model.add_slider("initial_bugs", 100, 10, 300)
stupid_model.add_controller_row()
stupid_model.add_slider("max_food_eat", 1.0, 0.1, 1.0)
stupid_model.add_slider("max_food_prod", 0.01, 0.01, 0.1)
stupid_model.add_controller_row()
stupid_model.add_slider("initialBugSizeMean", 1, 0, 10)
stupid_model.add_slider("initialBugSizeSD", 5, 0, 10)
stupid_model.histogram("grow_size", 0, 10, 5, (0, 0, 0))
stupid_model.line_chart(["current_bugs"], [(0, 0, 0)])
# Initialize tiles
for tile in model.tiles:
tile.info["food"] = 0.0
tile.color = (0, 0, 0)
def step(model):
# Food production
for tile in model.tiles:
food_prod = random.uniform(0, model.max_food_prod)
tile.info["food"] += food_prod
c = min(255, math.floor(tile.info["food"] * 255))
tile.color = (c, c, c)
# Move all agents
for agent in model.agents:
agent.step(model)
stupid_model = Model(
"StupidModel w. habitat cells and resources (stupid03)",
100,
100,
tile_size=5,
)
stupid_model.add_button("setup", setup)
stupid_model.add_button("step", step)
stupid_model.add_toggle_button("go", step)
run(stupid_model)
# Update plots
model.update_plots()
model.remove_destroyed_agents()
# TODO: Stop after 1000 iterations
if model.agent_count() == 0:
model.pause()
def close(model):
if file_handle:
file_handle.close()
stupid_model = Model(
"StupidModel w. population graph (stupid13)", 100, 100, tile_size=5
)
stupid_model.add_button("setup", setup)
stupid_model.add_button("step", step)
stupid_model.add_toggle_button("go", step)
stupid_model.add_controller_row()
stupid_model.add_slider("initial_bugs", 100, 10, 300)
stupid_model.add_controller_row()
stupid_model.add_slider("max_food_eat", 1.0, 0.1, 1.0)
stupid_model.add_controller_row()
stupid_model.add_slider("max_food_prod", 0.01, 0.01, 0.1)
stupid_model.histogram("grow_size", 0, 10, 5, (0, 0, 0))
stupid_model.line_chart(["current_bugs"], [(0, 0, 0)])
stupid_model.on_close(close)
run(stupid_model)
person.vaccinate(model)
exp_modifier += 0.2
break
file_handle = open("vaccine.csv", "a")
file_handle.write(
str(counter) + "," + str(model.Susceptible) + "," +
str(model.Infectious) + "," + str(model.Recovered) + "," +
str(model.Vaccinated) + "\n")
file_handle.close()
counter += 1
model.update_plots()
epidemic_model = Model("Epidemimodel", 100, 100)
epidemic_model.add_button("Setup", model_setup)
epidemic_model.add_button("Go", model_step, toggle=True)
epidemic_model.line_chart(
["Susceptible", "Infectious", "Recovered", "Vaccinated"],
[(0, 200, 0), (200, 0, 0), (0, 0, 200), (100, 100, 200)])
epidemic_model.bar_chart(
["Susceptible", "Infectious", "Recovered", "Vaccinated"], (200, 200, 200))
epidemic_model.add_checkbox("enable_groups")
epidemic_model.add_checkbox("Vaccine_exponential")
epidemic_model.add_controller_row()
epidemic_model.add_slider("Social_distance", 50, 0, 80)
epidemic_model.add_controller_row()
epidemic_model.add_slider("Infection_distance", 15, 0, 40)
def start_episode(self, state):
self.history = Counter()
self.model = Model(self.env)
self.plan = iter(()) # start with no plan
def model_setup(model):
model.reset()
model.Susceptible = 0
model.Infectious = 0
model.Recovered = 0
for person in range(100):
model.add_agent(Person())
def model_step(model):
for person in model.agents:
person.step(model)
model.update_plots()
epidemic_model = Model("Epidemimodel", 100, 100)
epidemic_model.add_button("Setup", model_setup)
epidemic_model.add_button("Go", model_step, toggle=True)
epidemic_model.line_chart(["Susceptible", "Infectious", "Recovered"],
[(0, 200, 0), (200, 0, 0), (0, 0, 200)])
epidemic_model.add_checkbox("enable_groups")
epidemic_model.add_controller_row()
epidemic_model.add_slider("social_distance", 50, 0, 80)
epidemic_model.add_controller_row()
epidemic_model.add_slider("infection_distance", 15, 0, 40)
run(epidemic_model)
def start_episode(self, state):
self.history = Counter()
self.model = Model(self.env)
def step(model):
# Food production
for tile in model.tiles:
food_prod = random.uniform(0, model.max_food_prod)
tile.info["food"] += food_prod
c = min(255, math.floor(tile.info["food"] * 255))
tile.color = (c, c, c)
# Move all agents
for agent in model.agents:
agent.step(model)
stupid_model = Model("StupidModel w. parameters and displays (stupid05)",
100,
100,
tile_size=5)
stupid_model.add_button("setup", setup)
stupid_model.add_button("step", step)
stupid_model.add_toggle_button("go", step)
stupid_model.add_controller_row()
stupid_model.add_slider(
"initial_bugs",
100,
10,
300,
)
stupid_model.add_controller_row()
stupid_model.add_slider("max_food_eat", 1.0, 0.1, 1.0)
stupid_model.add_controller_row()
stupid_model.add_slider("max_food_prod", 0.01, 0.01, 0.1)
# Write min, average and max bug size to file
file_handle.write(
str(bug_min) + " " + str(bug_mean) + " " + str(bug_max) + "\n")
# Update plots
model.update_plots()
def close(model):
if file_handle:
file_handle.close()
stupid_model = Model("StupidModel w. file output (stupid08)",
100,
100,
tile_size=5)
stupid_model.add_button("setup", setup)
stupid_model.add_button("step", step)
stupid_model.add_toggle_button("go", step)
stupid_model.add_controller_row()
stupid_model.add_slider("initial_bugs", 100, 10, 300)
stupid_model.add_controller_row()
stupid_model.add_slider("max_food_eat", 1.0, 0.1, 1.0)
stupid_model.add_controller_row()
stupid_model.add_slider("max_food_prod", 0.01, 0.01, 0.1)
stupid_model.histogram("grow_size", 0, 10, 5, (0, 0, 0))
stupid_model.on_close(close)
run(stupid_model)
self.size_to_color()
def step(self, model):
self.grow()
self.move()
def setup(model):
model.reset()
model.initial_bugs = 100
# Create and add agents
for i in range(int(model.initial_bugs)):
model.add_agent(Bug())
def step(model):
# Move all agents
for agent in model.agents:
agent.step(model)
stupid_model = Model("StupidModel w. Bug Growth (stupid02)",
100,
100,
tile_size=5)
stupid_model.add_button("setup", setup)
stupid_model.add_button("step", step)
stupid_model.add_toggle_button("go", step)
run(stupid_model)