class TestSingleGrid(unittest.TestCase):
def setUp(self):
self.space = SingleGrid(50, 50, False)
self.agents = []
for i, pos in enumerate(TEST_AGENTS_GRID):
a = MockAgent(i, None)
self.agents.append(a)
self.space.place_agent(a, pos)
def test_agent_positions(self):
"""
Ensure that the agents are all placed properly.
"""
for i, pos in enumerate(TEST_AGENTS_GRID):
a = self.agents[i]
assert a.pos == pos
def test_remove_agent(self):
for i, pos in enumerate(TEST_AGENTS_GRID):
a = self.agents[i]
assert a.pos == pos
assert self.space.grid[pos[0]][pos[1]] == a
self.space.remove_agent(a)
assert a.pos is None
assert self.space.grid[pos[0]][pos[1]] is None
def test_empty_cells(self):
if self.space.exists_empty_cells():
pytest.deprecated_call(self.space.find_empty)
for i, pos in enumerate(list(self.space.empties)):
a = MockAgent(-i, pos)
self.space.position_agent(a, x=pos[0], y=pos[1])
assert self.space.find_empty() is None
with self.assertRaises(Exception):
self.space.move_to_empty(a)
def move_agent(self):
agent_number = 0
initial_pos = TEST_AGENTS_GRID[agent_number]
final_pos = (7, 7)
_agent = self.agents[agent_number]
assert _agent.pos == initial_pos
assert self.space.grid[initial_pos[0]][initial_pos[1]] == _agent
assert self.space.grid[final_pos[0]][final_pos[1]] is None
self.space.move_agent(_agent, final_pos)
assert _agent.pos == final_pos
assert self.space.grid[initial_pos[0]][initial_pos[1]] is None
assert self.space.grid[final_pos[0]][final_pos[1]] == _agent
class TestSingleGrid(unittest.TestCase):
def setUp(self):
self.space = SingleGrid(50, 50, False)
self.agents = []
for i, pos in enumerate(TEST_AGENTS_GRID):
a = MockAgent(i, None)
self.agents.append(a)
self.space.place_agent(a, pos)
def test_agent_positions(self):
'''
Ensure that the agents are all placed properly.
'''
for i, pos in enumerate(TEST_AGENTS_GRID):
a = self.agents[i]
assert a.pos == pos
def test_remove_agent(self):
for i, pos in enumerate(TEST_AGENTS_GRID):
a = self.agents[i]
assert a.pos == pos
assert self.space.grid[pos[0]][pos[1]] == a
self.space.remove_agent(a)
assert a.pos is None
assert self.space.grid[pos[0]][pos[1]] is None
def move_agent(self):
agent_number = 0
initial_pos = TEST_AGENTS_GRID[agent_number]
final_pos = (7, 7)
_agent = self.agents[agent_number]
assert _agent.pos == initial_pos
assert self.space.grid[initial_pos[0]][initial_pos[1]] == _agent
assert self.space.grid[final_pos[0]][final_pos[1]] is None
self.space.move_agent(_agent, final_pos)
assert _agent.pos == final_pos
assert self.space.grid[initial_pos[0]][initial_pos[1]] is None
assert self.space.grid[final_pos[0]][final_pos[1]] == _agent
class TestSingleGrid(unittest.TestCase):
def setUp(self):
self.space = SingleGrid(50, 50, False)
self.agents = []
for i, pos in enumerate(TEST_AGENTS_GRID):
a = MockAgent(i, None)
self.agents.append(a)
self.space.place_agent(a, pos)
def test_agent_positions(self):
'''
Ensure that the agents are all placed properly.
'''
for i, pos in enumerate(TEST_AGENTS_GRID):
a = self.agents[i]
assert a.pos == pos
def test_remove_agent(self):
for i, pos in enumerate(TEST_AGENTS_GRID):
a = self.agents[i]
assert a.pos == pos
assert self.space.grid[pos[0]][pos[1]] == a
self.space.remove_agent(a)
assert a.pos is None
assert self.space.grid[pos[0]][pos[1]] is None
def move_agent(self):
agent_number = 0
initial_pos = TEST_AGENTS_GRID[agent_number]
final_pos = (7, 7)
_agent = self.agents[agent_number]
assert _agent.pos == initial_pos
assert self.space.grid[initial_pos[0]][initial_pos[1]] == _agent
assert self.space.grid[final_pos[0]][final_pos[1]] is None
self.space.move_agent(_agent, final_pos)
assert _agent.pos == final_pos
assert self.space.grid[initial_pos[0]][initial_pos[1]] is None
assert self.space.grid[final_pos[0]][final_pos[1]] == _agent
class EvacuationModel(Model):
"""
This is a simulation of a crowd evacuation from a building.
Several variables are taken into account: the knowledge of the emergency exits, the age and weight of the agents
and the presence of stewards that can guide agents toward the emergency exits.
Agents have different strategies to escape the building such as taking the shortest path to an exit or a random one.
The goal is to study which combinations of agent types are more likely to escape the building and save themselves and
how the amount of casualties varies with respect to the different variables.
"""
def __init__(self,
N=10,
K=0,
width=50,
height=50,
fire_x=1,
fire_y=1,
civil_info_exchange=True):
self.num_civilians = N
self.num_stewards = K
self.civil_info_exchange = civil_info_exchange
self.fire_initial_pos = (fire_x, fire_y)
self.warning_UI = ""
self.agents_alive = N + K # Agents alive and inside the building
self.agents_saved = [] # Agents that managed to get out
self.agents_killed = [] # Agents that perished during the evacuation
self.grid = SingleGrid(height, width, False)
self.graph = None # General graph representing walkable terrain
self.schedule = RandomActivation(
self) # Every tick, agents move in a different random order
# Create exits
self.pos_exits = [(0, 5), (0, 25), (0, 45)]
for i in range(3):
self.pos_exits.append((self.grid.width - 1, 14 + i))
self.draw_environment(self.pos_exits)
self.graph = path_finding.create_graph(self)
# Define data collector
model_collector = {
"Agents killed": lambda killed: len(self.agents_killed),
"Agents saved": lambda saved: len(self.agents_saved)
}
for exit_pos in self.pos_exits:
title = "Exit {}".format(exit_pos)
model_collector[title] = partial(count_agents_saved, exit_pos)
self.datacollector = DataCollector(model_reporters=model_collector)
# Create fire
# for pos in self.fire_initial_pos: # Only 1 source of fire since we are setting it from UI
x, y = self.fire_initial_pos
if not self.is_inside_square((x, y), (0, 29),
(25, 39)) and not self.is_inside_square(
(x, y), (0, 10), (25, 20)):
pos = self.fire_initial_pos
else:
pos = (1, 1)
self.warning_UI = "<b>WARNING:</b> Sorry but the position of the fire is outside of the building, " \
"change the setting and click reset simulation."
fire_agent = FireAgent(pos, self)
self.schedule.add(fire_agent)
self.grid.place_agent(fire_agent, pos)
# Create civilian agents
for i in range(self.num_civilians):
# a civilian agent will know at least the main entrance to the building
known_exits = self.pos_exits[-3:]
a = CivilianAgent(i, self, known_exits)
self.schedule.add(a)
# Add the agent to a random grid cell
while True:
# pick the random coordinate
x = self.random.randrange(1, self.grid.width - 1)
y = self.random.randrange(1, self.grid.height - 1)
# check if the point is empty and inside of the building
if self.grid.is_cell_empty((x, y)) and not self.is_inside_square((x, y), (0, 29), (25, 39)) \
and not self.is_inside_square((x, y), (0, 10), (25, 20)):
break
self.grid.place_agent(a, (x, y))
# Create steward agents
for i in range(self.num_civilians,
self.num_civilians + self.num_stewards):
# a steward agent will know all exits.
known_exits = self.pos_exits
a = StewardAgent(i, self, known_exits)
self.schedule.add(a)
# Add the agent to a random grid cell
while True:
# pick the random coordinate
x = self.random.randrange(1, self.grid.width - 1)
y = self.random.randrange(1, self.grid.height - 1)
# check if the point is empty and inside of the building
if self.grid.is_cell_empty((x, y)) and not self.is_inside_square((x, y), (0, 29), (25, 39)) \
and not self.is_inside_square((x, y), (0, 10), (25, 20)):
break
self.grid.place_agent(a, (x, y))
self.running = True # Set this to false when we want to finish simulation (e.g. all agents are out of building)
self.datacollector.collect(self)
@staticmethod
def is_inside_square(point, bottom_left, top_right):
return bottom_left[0] <= point[0] <= top_right[0] and bottom_left[
1] <= point[1] <= top_right[1]
def step(self):
self.schedule.step()
# collect data
self.datacollector.collect(self)
# Halt if no more agents in the building
if self.count_agents(self) == 0:
self.running = False
def remove_agent(self, agent, reason, **kwargs):
"""
Removes an agent from the simulation. Depending on the reason it can be
Args:
agent (Agent):
reason (Reasons):
Returns:
None
"""
if reason == Reasons.SAVED:
self.agents_saved.append(agent)
elif reason == Reasons.KILLED_BY_FIRE:
self.agents_killed.append(agent)
self.agents_alive -= 1
self.schedule.remove(agent)
self.grid.remove_agent(agent)
def draw_environment(self, exits=None):
length_E = int(self.grid.height /
5) # length of the vertical segments of the E
depth_E = int(self.grid.width /
2) # length of the horizontal segments of the E
for i in range(3):
start = max(0, 2 * i * length_E)
self.draw_wall((0, start), (0, start + length_E - 1))
for i in range(2):
start = 2 * i * length_E + length_E
self.draw_wall((depth_E, start), (depth_E, start + length_E - 1))
# Horizontal lines of the E (BB)
aux_y_coord = [
length_E, 2 * length_E, 3 * length_E - 1, 4 * length_E - 1
]
for y in aux_y_coord:
self.draw_wall((0, y), (depth_E, y))
top_left_corner = (0, self.grid.height - 1)
top_right_corner = (self.grid.width - 1, self.grid.height - 1)
bottom_right_corner = (self.grid.width - 1, 0)
# Draw long contour lines E
self.draw_wall((0, 0), bottom_right_corner)
self.draw_wall(top_left_corner, top_right_corner)
self.draw_wall(bottom_right_corner, top_right_corner)
# Draw exits
self.draw_exits(exits)
def draw_wall(self, start, end):
"""
Draws a line that goes from start point to end point.
Args:
start (List): Coordinates of line's starting point
end (List): Coordinates of line's end point
Returns:
None
"""
diff_x, diff_y = np.subtract(end, start)
wall_coordinates = np.asarray(start)
if self.grid.is_cell_empty(wall_coordinates.tolist()):
w = WallAgent(wall_coordinates.tolist(), self)
self.grid.place_agent(w, wall_coordinates.tolist())
while diff_x != 0 or diff_y != 0:
if abs(diff_x) == abs(diff_y):
# diagonal wall
wall_coordinates[0] += np.sign(diff_x)
wall_coordinates[1] += np.sign(diff_y)
diff_x -= 1
diff_y -= 1
elif abs(diff_x) < abs(diff_y):
# wall built in y dimension
wall_coordinates[1] += np.sign(diff_y)
diff_y -= 1
else:
# wall built in x dimension
wall_coordinates[0] += np.sign(diff_x)
diff_x -= 1
if self.grid.is_cell_empty(wall_coordinates.tolist()):
w = WallAgent(wall_coordinates.tolist(), self)
self.grid.place_agent(w, wall_coordinates.tolist())
def draw_exits(self, exits_list):
for ext in exits_list:
e = ExitAgent(ext, self)
if not self.grid.is_cell_empty(ext):
# Only walls should exist in the grid at this time, so no need to remove it from scheduler
agent = self.grid.get_cell_list_contents(ext)
self.grid.remove_agent(agent[0])
# Place exit
self.schedule.add(e)
self.grid.place_agent(e, ext)
def spread_fire(self, fire_agent):
fire_neighbors = self.grid.get_neighborhood(fire_agent.pos,
moore=True,
include_center=False)
for grid_space in fire_neighbors:
if self.grid.is_cell_empty(grid_space):
# Create new fire agent and add it to grid and scheduler
new_fire_agent = FireAgent(grid_space, self)
self.schedule.add(new_fire_agent)
self.grid.place_agent(new_fire_agent, grid_space)
else:
# If human agents, eliminate them and spread anyway
agent = self.grid.get_cell_list_contents(grid_space)[0]
if isinstance(agent, (CivilianAgent, StewardAgent)):
new_fire_agent = FireAgent(grid_space, self)
self.remove_agent(agent, Reasons.KILLED_BY_FIRE)
self.schedule.add(new_fire_agent)
self.grid.place_agent(new_fire_agent, grid_space)
@staticmethod
def count_agents(model):
"""
Helper method to count agents alive and still in the building.
"""
count = 0
for agent in model.schedule.agents:
agent_type = type(agent)
if (agent_type == CivilianAgent) or (agent_type == StewardAgent):
count += 1
return count
class Anthill(Model):
def __init__(self):
self.grid = SingleGrid(WIDTH, HEIGHT, False)
self.schedule = RandomActivation(self)
self.running = True
self.internalrate = 0.2
self.ant_id = 1
self.tau = np.zeros((WIDTH, HEIGHT))
self.datacollector = DataCollector({
"Total number of Ants":
lambda m: self.get_total_ants_number(),
"mean tau":
lambda m: self.evaluation1(),
"sigma":
lambda m: self.evaluation2(),
"sigma*":
lambda m: self.evaluation3(),
})
# List containing all coordinates of the boundary, initial ants location and brood location
self.bound_vals = []
self.neigh_bound = []
self.datacollector.collect(self)
for i in range(WIDTH):
for j in range(HEIGHT):
if i == 0 or j == 0 or i == WIDTH - 1 or j == HEIGHT - 1:
self.bound_vals.append((i, j))
if i == 1 or i == WIDTH - 2 or j == 1 or j == HEIGHT - 2:
self.neigh_bound.append((i, j))
# Make a Fence boundary
b = 0
for h in self.bound_vals:
br = Fence(b, self)
self.grid.place_agent(br, (h[0], h[1]))
b += 1
def step(self):
'''Advance the model by one step.'''
# Add new ants into the internal area ont he boundary
for xy in self.neigh_bound:
# Add with probability internal rate and if the cell is empty
if self.random.uniform(
0, 1) < self.internalrate and self.grid.is_cell_empty(
xy) == True:
a = Ant(self.ant_id, self)
self.schedule.add(a)
self.grid.place_agent(a, xy)
self.ant_id += 1
# Move the ants
self.schedule.step()
self.datacollector.collect(self)
# Remove all ants on bounary
for (agents, i, j) in self.grid.coord_iter():
if (i, j) in self.neigh_bound and type(agents) is Ant:
self.grid.remove_agent(agents)
self.schedule.remove(agents)
data_tau.append(self.mean_tau_ant)
data_sigma.append(np.sqrt(self.sigma))
data_sigmastar.append(self.sigmastar)
if len(data_sigmastar) > 20:
if abs(data_sigmastar[-2] - data_sigmastar[-1]) < 0.0000001 or len(
data_sigmastar) == 2000:
try:
# TAU
with open("results/m1_tau_5.pkl", 'rb') as f:
tau_old = pickle.load(f)
tau_old[int(len(tau_old) + 1)] = data_tau
f.close()
pickle.dump(tau_old, open("results/m1_tau_5.pkl", 'wb'))
except:
pickle.dump({1: data_tau},
open("results/m1_tau_5.pkl", 'wb'))
try:
# SIGMA
with open("results/m1_sigma_5.pkl", 'rb') as f:
sigma_old = pickle.load(f)
sigma_old[int(len(sigma_old) + 1)] = data_sigma
f.close()
pickle.dump(sigma_old, open("results/m1_sigma_5.pkl",
'wb'))
except:
pickle.dump({1: data_sigma},
open("results/m1_sigma_5.pkl", 'wb'))
try:
# SIGMASTAR
with open("results/m1_sigmastar_5.pkl", 'rb') as f:
sigmastar_old = pickle.load(f)
sigmastar_old[int(len(sigmastar_old) +
1)] = data_sigmastar
f.close()
pickle.dump(sigmastar_old,
open("results/m1_sigmastar_5.pkl", 'wb'))
except:
pickle.dump({1: data_sigmastar},
open("results/m1_sigmastar_5.pkl", 'wb'))
try:
# MATRIX
with open("results/m1_matrix_5.pkl", 'rb') as f:
matrix_old = pickle.load(f)
matrix_old[int(len(matrix_old) + 1)] = self.tau
f.close()
pickle.dump(matrix_old,
open("results/m1_matrix_5.pkl", 'wb'))
except:
pickle.dump({1: self.tau},
open("results/m1_matrix_5.pkl", 'wb'))
print(
"_______________________________________________________________________"
)
print("DONE")
self.running = False
# with open("tau2_new.txt", "a") as myfile:
# myfile.write(str(self.mean_tau_ant) + '\n')
# with open("sigma2_new.txt", "a") as myfile:
# myfile.write(str(np.sqrt(self.sigma)) + '\n')
# with open("datasigmastar2_new.txt","a") as myfile:
# myfile.write(str(self.sigmastar) + "\n")
def get_total_ants_number(self):
total_ants = 0
for (agents, _, _) in self.grid.coord_iter():
if type(agents) is Ant:
total_ants += 1
return total_ants
def evaluation1(self):
##creat a empty grid to store currently information
total_ants = np.zeros((WIDTH, HEIGHT))
## count the number of currently information
for (agents, i, j) in self.grid.coord_iter():
if type(agents) is Ant:
total_ants[i][j] = 1
else:
total_ants[i][j] = 0
##update the tau
self.tau = self.tau + total_ants
##calcualte the mean tau
self.mean_tau_ant = self.tau.sum() / ((WIDTH - 2)**2)
return self.mean_tau_ant
def evaluation2(self):
## we need to minus the mean tau so we need to ensure the result of boundary is zero
## so we let the bounday equal mean_tau_ant in this way the (tau-mean_tau_ant) is zero of boundary
for site in self.bound_vals:
self.tau[site[0]][site[1]] = self.mean_tau_ant
## calculate the sigmaa
self.sigma = ((self.tau - self.mean_tau_ant)**2).sum() / (
(WIDTH - 2)**2)
## rechange the boundaryy
for site in self.bound_vals:
self.tau[site[0]][site[1]] = 0
return np.sqrt(self.sigma)
def evaluation3(self):
## calculate the sigmastar
self.sigmastar = np.sqrt(self.sigma) / self.mean_tau_ant
return self.sigmastar
class modelSim(Model):
"""
details of the world
introduce time is when animal agents first get introduced into the wrold
disp_rate is the dispersal rate for experiment 3
dist is perceptual strength for animals if fixed
det is decision determinacy of animals if fixed
cog_fixed determines if cognition of animals is fixed to particular values or is allowed to evolve
if skip_300 is True, patchiness values are not calculated for the first 300 steps-- this makes the model run faster
collect_cog_dist creates a seperate dataframe for all cognition values for agents at every timestep
if evolve_disp is true, dispersion rate of plants is free to evolve
"""
def __init__(self, introduce_time, disp_rate, dist, det, cog_fixed = False, \
skip_300 = True, collect_cog_dist = False, evolve_disp = False):
self.skip_300 = skip_300
self.cog_fixed = cog_fixed
self.evolve_disp = evolve_disp
self.collect_cog_dist = collect_cog_dist
self.dist = dist
self.det = det
self.disp_rate = disp_rate
self.intro_time = introduce_time
(self.a1num, self.a2num) = (20, 20)
self.schedule = RandomActivation(
self) # agents take a step in random order
self.grid = SingleGrid(
200, 200,
True) # the world is a grid with specified height and width
self.initialize_perception()
disp = np.power(self.disp_rate, range(0, 100))
self.disp = disp / sum(disp)
self.grid_ind = np.indices((200, 200))
positions = np.maximum(abs(100 - self.grid_ind[0]),
abs(100 - self.grid_ind[1]))
self.positions = np.minimum(positions, 200 - positions)
self.agentgrid = np.zeros(
(self.grid.width, self.grid.height
)) # allows for calculation of patchiness of both agents
self.coggrid = np.full(
(self.nCogPar, self.grid.width, self.grid.height), 101.0)
self.dispgrid = np.full((2, self.grid.width, self.grid.height), 101.0)
self.age = []
(self.nstep, self.unique_id, self.reprod, self.food, self.death,
self.combat) = (0, 0, 0, 0, 0, 0)
self.cmap = colors.ListedColormap([
'midnightblue', 'mediumseagreen', 'white', 'white', 'white',
'white', 'white'
]) #'yellow', 'orange', 'red', 'brown'])
bounds = [0, 1, 2, 3, 4, 5, 6, 7]
self.norm = colors.BoundaryNorm(bounds, self.cmap.N)
self.expect_NN = []
self.NN = [5, 10]
for i in self.NN:
self.expect_NN.append(
(math.factorial(2 * i) * i) / (2**i * math.factorial(i))**2)
grid_ind_food = np.indices((21, 21))
positions_food = np.maximum(abs(10 - grid_ind_food[0]),
abs(10 - grid_ind_food[1]))
self.positions_food = np.minimum(positions_food, 21 - positions_food)
if self.collect_cog_dist:
self.cog_dist_dist = pd.DataFrame(columns=[])
self.cog_dist_det = pd.DataFrame(columns=[])
for i in range(self.a1num): # initiate a1 agents at random locations
self.introduce_agents("A1")
self.nA1 = self.a1num
self.nA2 = 0
# self.agent_steps = {}
def initialize_perception(self):
self.history = pd.DataFrame(columns=[
"nA1", "nA2", "age", "LIP5", "LIP10", "LIPanim5", "LIPanim10",
"Morsita5", "Morsita10", "Morsitaanim5", "Morsitaanim10", "NN5",
"NN10", "NNanim5", "NNanim10", "reprod", "food", "death", "combat",
"dist", "det", "dist_lower", "det_lower", "dist_upper",
"det_upper", "dist_ci", "det_ci"
])
self.nCogPar = 2
(self.start_energy, self.eat_energy, self.tire_energy, self.reproduction_energy, self.cognition_energy) \
= (10, 5, 3, 20, 1)
def introduce_agents(self, which_agent):
x = random.randrange(self.grid.width)
y = random.randrange(self.grid.height)
if which_agent == "A1":
if self.grid.is_cell_empty((x, y)):
a = A1(self.unique_id, self, self.start_energy, disp_rate=0)
self.unique_id += 1
self.grid.position_agent(a, x, y)
self.schedule.add(a)
self.agentgrid[x][y] = 1
else:
self.introduce_agents(which_agent)
elif which_agent == "A2":
if self.cog_fixed:
c = (self.dist, self.det)
else:
c = tuple([0] * self.nCogPar)
a = A2(self.unique_id,
self,
self.start_energy,
cognition=c,
disp_rate=0)
self.unique_id += 1
if self.agentgrid[x][y] == 1:
die = self.grid.get_cell_list_contents([(x, y)])[0]
die.dead = True
self.grid.remove_agent(die)
self.schedule.remove(die)
self.grid.place_agent(a, (x, y))
self.schedule.add(a)
self.agentgrid[x][y] = 2
self.coggrid[:, x, y] = c
elif self.agentgrid[x][y] == 0:
self.grid.place_agent(a, (x, y))
self.schedule.add(a)
self.agentgrid[x][y] = 2
self.coggrid[:, x, y] = c
def flatten_(self, n, grid, full_grid=False, mean=True, range_=False):
if full_grid:
return (grid[n].flatten())
i = grid[n].flatten()
if mean:
i = np.delete(i, np.where(i == 101))
if len(i) == 0:
# if range_:
return ([0] * 4)
#else:
# return(0)
if range_:
if self.cog_fixed:
return ([np.mean(i)] * 4)
return (np.concatenate(
([np.mean(i)], np.percentile(i, [2.5, 97.5]),
self.calculate_ci(i))))
return ([np.mean(i), 0, 0, 0])
else:
return (i)
def calculate_ci(self, data):
if np.min(data) == np.max(data):
return ([0.0])
return ([
np.mean(data) - st.t.interval(
0.95, len(data) - 1, loc=np.mean(data), scale=st.sem(data))[0]
])
def return_zero(self, num, denom):
if self.nstep == 1:
# print("whaaat")
return (0)
if denom == "old_nA2":
denom = self.history["nA2"][self.nstep - 2]
if denom == 0.0:
return 0
return (num / denom)
def nearest_neighbor(self, agent): # fix this later
if agent == "a1":
x = np.argwhere(self.agentgrid == 1)
if len(x) <= 10:
return ([-1] * len(self.NN))
elif len(x) > 39990:
return ([0.97, 0.99])
# if self.nstep<300 and self.skip_300:
# return([-1,-1] )
else:
x = np.argwhere(self.agentgrid == 2)
if len(x) <= 10:
return ([-1] * len(self.NN))
density = len(x) / (self.grid.width)**2
expect_NN_ = self.expect_NN
expect_dist = np.array(expect_NN_) / (density**0.5)
distances = [0, 0]
for i in x:
distx = abs(x[:, 0] - i[0])
distx[distx > 100] = 200 - distx[distx > 100]
disty = abs(x[:, 1] - i[1])
disty[disty > 100] = 200 - disty[disty > 100]
dist = (distx**2 + disty**2)**0.5
distances[0] += (np.partition(dist, 5)[5])
distances[1] += (np.partition(dist, 10)[10])
mean_dist = np.array(distances) / len(x)
out = mean_dist / expect_dist
return (out)
def quadrant_patch(
self, agent
): # function to calculate the patchiness index of agents at every step
if agent == "a1":
x = self.agentgrid == 1
else:
x = self.agentgrid == 2
gsize = np.array([5, 10])
gnum = 200 / gsize
qcs = []
for i in range(2):
x_ = x.reshape(int(gnum[i]), gsize[i], int(gnum[i]),
gsize[i]).sum(1).sum(2)
mean = np.mean(x_)
var = np.var(x_)
if mean == 0.0:
return ([-1] * 4)
lip = 1 + (var - mean) / (mean**2)
morsita = np.sum(x) * ((np.sum(np.power(x_, 2)) - np.sum(x_)) /
(np.sum(x_)**2 - np.sum(x_)))
qcs += [lip, morsita]
return (qcs)
def l_function(self, agent):
if agent == "a1":
x = np.argwhere(self.agentgrid == 1)
else:
x = np.argwhere(self.agentgrid == 2)
if len(x) == 0:
return (-1)
distances = np.array([])
for i in x:
distx = abs(x[:, 0] - i[0])
distx[distx > 100] = 200 - distx[distx > 100]
disty = abs(x[:, 1] - i[1])
disty[disty > 100] = 200 - disty[disty > 100]
dist = (distx**2 + disty**2)**0.5
distances = np.concatenate((distances, dist[dist != 0]))
l = np.array([])
for i in np.arange(5, 51, 5):
l = np.append(l, sum(distances < i))
k = (l * 200**2) / (len(x)**2)
l = (k / math.pi)**0.5
return (abs(l - np.arange(5, 51, 5)))
def collect_hist(self):
if self.nstep < 300 and self.skip_300:
NNcalc = [-1, -1] #self.nearest_neighbor("a1")
NNanimcalc = [-1, -1] #self.nearest_neighbor("a2")
else:
NNcalc = self.nearest_neighbor("a1")
NNanimcalc = self.nearest_neighbor("a2")
quadrantcalc = self.quadrant_patch("a1")
quadrantanimcalc = self.quadrant_patch("a2")
dist_values = self.flatten_(0,
grid=self.coggrid,
mean=True,
range_=False)
det_values = self.flatten_(1,
grid=self.coggrid,
mean=True,
range_=False)
# l_f = 0#self.l_function("a1")
dat = {
"nA1": self.nA1,
"nA2": self.nA2,
"age": self.return_zero(sum(self.age), self.nA2),
"LIP5": quadrantcalc[0],
"LIP10": quadrantcalc[2],
"LIPanim5": quadrantanimcalc[0],
"LIPanim10": quadrantanimcalc[2],
"Morsita5": quadrantcalc[1],
"Morsita10": quadrantcalc[3],
"Morsitaanim5": quadrantanimcalc[1],
"Morsitaanim10": quadrantanimcalc[3],
"NN5": NNcalc[0],
"NN10": NNcalc[1],
"NNanim5": NNanimcalc[0],
"NNanim10":
NNanimcalc[1], #"l_ripley" : l_f,# self.nearest_neighbor("a2"),
"reprod": self.return_zero(self.reprod, "old_nA2"),
"food": self.return_zero(self.food, self.nA2),
"death": self.return_zero(self.death, "old_nA2"),
"combat": self.return_zero(self.combat, "old_nA2"),
"dist": dist_values[0],
"det": det_values[0],
"dist_lower": dist_values[1],
"det_lower": det_values[1],
"dist_upper": dist_values[2],
"det_upper": det_values[2],
"dist_ci": dist_values[3],
"det_ci": det_values[3],
"disp_a1": self.flatten_(0, grid=self.dispgrid)[0],
"disp_a2": self.flatten_(1, grid=self.dispgrid)[0]
}
self.history = self.history.append(dat, ignore_index=True)
self.age = []
(self.reprod, self.food, self.death, self.combat) = (0, 0, 0, 0)
if self.collect_cog_dist:
if (self.nstep % 10) == 0:
self.cog_dist_dist[str(self.nstep - 1)] = self.flatten_(
0, grid=self.coggrid, full_grid=True, mean=False)
self.cog_dist_det[str(self.nstep - 1)] = self.flatten_(
1, grid=self.coggrid, full_grid=True, mean=False)
def step(self):
self.nstep += 1 # step counter
if self.nstep == self.intro_time:
for i in range(self.a2num):
self.introduce_agents("A2")
self.schedule.step()
self.nA1 = np.sum(self.agentgrid == 1)
self.nA2 = np.sum(self.agentgrid == 2)
self.collect_hist()
if self.nstep % 10 == 0:
sys.stdout.write((str(self.nstep) + " " + str(self.nA1) + " " +
str(self.nA2) + "\n"))
def visualize(self):
f, ax = plt.subplots(1)
self.agentgrid = self.agentgrid.astype(int)
ax.imshow(self.agentgrid,
interpolation='nearest',
cmap=self.cmap,
norm=self.norm)
# plt.axis("off")
return (f)
class SDGrid(Model):
''' Model class for iterated, spatial social dilemma model. '''
schedule_types = {"Sequential": BaseScheduler,
"Random": RandomActivation,
"Simultaneous": SimultaneousActivation}
# This dictionary holds the payoff for this agent,
# keyed on: (my_move, other_move)
#NOTE: Payoffs must be given by the user in the format below as a dict object.
def __init__(self, height=0, width=0, schedule_type="Random", payoffs=None, seed=2514, p = .1, implement = "Epstein", num_RL =500, ep_length=1):
'''
Create a new Spatial Prisoners' Dilemma Model.
Args:
height, width: Grid size. There will be one agent per grid cell.
schedule_type: Can be "Sequential", "Random", or "Simultaneous".
Determines the agent activation regime.
payoffs: (required) Dictionary of (move, neighbor_move) payoffs.
'''
#Set default grid size if none inputted by user
if height == 0:
h = 50
else:
h = height
if width == 0:
w = 50
else:
w = width
assert height or width < 0, "Grid heigth and width must be positive numbers."
if payoffs:
self.payoff = payoffs
else:
self.payoff = {(C, C): 5,
(C, D): -5,
(D, C): 6,
(D, D): -6}
self.grid = SingleGrid(h, w, torus=True)
self.schedule_type = schedule_type
self.schedule = self.schedule_types[self.schedule_type](self)
self.implement = implement
self.ep_length = ep_length
self.num_RL = num_RL
#FIXME: THis is a bandaid fix for MESA's loop bug (see trello for SD ABM):
self.kill_list = []
self.fertile_agents = []
if self.implement == "Epstein":
leave_empty = np.random.choice(a = [False, True], size = (width, height), p = [p, 1-p])
else:
pass
# Create agents: automatically populates agents and grid;
count = 0
for x in range(width):
for y in range(height):
if implement == "Epstein":
if leave_empty[x, y]:
continue
else:
agent = SDAgent(count, (x, y), self)
count +=1
else:
agent = SDAgent(count, (x, y), self)
count +=1
self.grid.place_agent(agent, (x, y))
self.schedule.add(agent)
########################################################################################################################
#FIXME: may need to make an unique id for agents to do this correctly
#FIXME: this will have to be generalized later for when multipe batches of episodes are being run
########################################################################################################################
# learners = np.random.choice([self.schedule._agents], self.num_RL)
# #switch them to learn mode
# for agent in learners:
# agent.learn_mode = True
# TODO: Make data collection easier for user; need to do same in BWT / Cartel
self.datacollector = DataCollector(model_reporters={
"Learning Cooperating_Agents":
lambda m: (len([a for a in m.schedule.agents if a.move == C and a.unique_id == 1] )),
"Learning Defecting_Agents":
lambda n: (len([b for b in n.schedule.agents if b.move == D and b.unique_id == 1] ))
})
self.running = True
self.datacollector.collect(self)
def step(self):
self.schedule.step()
for agent in self.schedule.agents:
if agent.unique_id == 1:
agent.learn = True
#print('agent 1 has learn set to {}'.format(agent.learn))
agent.update_policy()
# collect data
self.datacollector.collect(self)
# self.purge()
# self.replicate_agents()
#if (self.schedule.time % self.ep_length == 0) and (self.schedule.time > 0):
#print(self.schedule.time)
# learners = random.sample(self.schedule.agents, self.num_RL)
# for agent in learners:
# agent.learn = True
# for agent in learners:
# agent.update_policy()
# agent.learn = False
# if agent == learners[-1]:
# print("################################# Update finished #################################")
def replicate_agents(self):
if self.fertile_agents is not None:
for agent in self.fertile_agents:
if agent.pos is not None:
try:
agent.replicate()
except ValueError:
#print("Caught a bad egg, boss!")
continue
def purge(self):
if len(self.kill_list)>0:
for agent in self.kill_list:
self.grid.remove_agent(agent)
self.schedule.remove(agent)
self.kill_list = []
else:
pass
def run(self, n):
''' Run the model for n steps. '''
for _ in range(n):
self.step()
class RoadModel(Model):
"""
A model with a number of cars, Nagel-Schreckenberg
"""
def __init__(self, N, length=100, lanes=1, timer=3):
self.num_agents = N
self.grid = SingleGrid(length, lanes, torus=True)
model_stages = [
"acceleration", "braking", "randomisation", "move", "delete"
]
self.schedule = StagedActivation(self, stage_list=model_stages)
# Create agent
for i in range(self.num_agents):
agent = CarAgent(i, self, False)
# Add to schedule
self.schedule.add(agent)
# Add to grid (randomly)
self.grid.position_agent(agent)
# Add the traffic light
self.traffic_light = TrafficLight(0, self, timer, 20, 20)
self.average_velocity = CarAgent.init_velocity
self.datacollector = DataCollector(agent_reporters={
"Position": "pos",
"Velocity": "velocity"
},
model_reporters={
"Average Velocity":
"average_velocity",
"Amount of cars": "agent_count",
"On Ramp Queue":
get_on_ramp_queue,
"Waiting Queue":
get_waiting_queue
})
self.running = True
def step(self):
"""
The model takes a new step and updates
"""
# Calculate amount of agents
self.agent_count = len(self.schedule.agents)
# Calculate average velocity
self.average_velocity = np.mean(
[a.velocity for a in self.schedule.agents])
# Collect data
self.datacollector.collect(self)
# Run a step of the traffic light
self.traffic_light.step()
# Run next step
self.schedule.step()
def add_agent(self, label, x_corr):
"""
Adds an agent to the scheduler and model on a particular coordinate
:param label: The label of the agents that gets created
:param x_corr: The x-coordinate of where the agent will be spawned
"""
# Create agent
agent = CarAgent(label, self, True)
# Add to schedule
self.schedule.add(agent)
# Add to grid on a certain position
self.grid.position_agent(agent, x_corr, 0)
def delete_agent(self, agent):
"""
Deletes an agent from the scheduler and model
:param agent: The agents that gets deleted
"""
# remove from schedule
self.schedule.remove(agent)
# remove from grid
self.grid.remove_agent(agent)
class GTModel(Model):
def __init__(self, debug, size, i_n_agents, i_strategy, i_energy,
child_location, movement, k, T, M, p, d, strategies_to_count,
count_tolerance, mutation_type, death_threshold, n_groups):
self.grid = SingleGrid(size, size, torus=True)
self.schedule = RandomActivation(self)
self.running = True
self.debug = debug
self.size = size
self.agent_idx = 0
self.i_energy = i_energy
# Payoff matrix in the form (my_move, op_move) : my_reward
self.payoff = {
('C', 'C'): 2,
('C', 'D'): -3,
('D', 'C'): 3,
('D', 'D'): -1,
}
# Constant for max population control (cost of surviving)
self.k = k
# Constant for controlling dying of old age
self.M = M
# Minimum lifespan
self.T = T
# Minimum energy level to reproduce
self.p = p
# Mutation "amplitude"
self.d = d
# Whether to spawn children near parents or randomly
self.child_location = child_location
# Specify the type of movement allowed for the agents
self.movement = movement
# Specify how the agents mutate
self.mutation_type = mutation_type
# The minimum total_energy needed for an agent to survive
self.death_threshold = death_threshold
# Vars regarding which strategies to look for
self.strategies_to_count = strategies_to_count
self.count_tolerance = count_tolerance
# Add agents (one agent per cell)
all_coords = [(x, y) for x in range(size) for y in range(size)]
agent_coords = self.random.sample(all_coords, i_n_agents)
for _ in range(i_n_agents):
group_idx = (None if n_groups is None else self.random.choice(
range(n_groups)))
agent = GTAgent(self.agent_idx, group_idx, self, i_strategy.copy(),
i_energy)
self.agent_idx += 1
self.schedule.add(agent)
self.grid.place_agent(agent, agent_coords.pop())
# Collect data
self.datacollector = DataCollector(
model_reporters={
**{
'strategies': get_strategies,
'n_agents': total_n_agents,
'avg_agent_age': avg_agent_age,
'n_friendlier': n_friendlier,
'n_aggressive': n_aggressive,
'perc_cooperative_actions': perc_cooperative_actions,
'n_neighbors': n_neighbor_measure,
'avg_delta_energy': avg_delta_energy,
'perc_CC': perc_CC_interactions,
'lin_fit_NC': coop_per_neig,
'lin_fit_NC_intc': coop_per_neig_intc,
},
**{
label: strategy_counter_factory(strategy, count_tolerance)
for label, strategy in strategies_to_count.items()
}
})
def alpha(self):
# Return the cost of surviving, alpha
DC = self.payoff[('D', 'C')]
CC = self.payoff[('C', 'C')]
N = len(self.schedule.agents)
return self.k + 4 * (DC + CC) * N / (self.size * self.size)
def time_to_die(self, agent):
# There is a chance every iteration to die of old age: (A - T) / M
# There is a 100% to die if the agents total energy reaches 0
return (agent.total_energy < self.death_threshold
or self.random.random() < (agent.age - self.T) / self.M)
def get_child_location(self, agent):
if self.child_location == 'global':
return self.random.choice(sorted(self.grid.empties))
elif self.child_location == 'local':
# Iterate over the radius, starting at 1 to find empty cells
for rad in range(1, int(self.size / 2)):
possible_steps = [
cell for cell in self.grid.get_neighborhood(
agent.pos,
moore=False,
include_center=False,
radius=rad,
) if self.grid.is_cell_empty(cell)
]
if possible_steps:
return self.random.choice(possible_steps)
# If no free cells in radius size/2 pick a random empty cell
return self.random.choice(sorted(self.grid.empties))
def maybe_mutate(self, agent):
# Mutate by adding a random d to individual Pi's
if self.mutation_type == 'stochastic':
# Copy the damn list
new_strategy = agent.strategy.copy()
# There is a 20% chance of mutation
if self.random.random() < 0.2:
# Each Pi is mutated uniformly by [-d, d]
for i in range(4):
mutation = self.random.uniform(-self.d, self.d)
new_val = new_strategy[i] + mutation
# Keep probabilities in [0, 1]
new_val = (0 if new_val < 0 else
1 if new_val > 1 else new_val)
new_strategy[i] = new_val
# Mutate by choosing a random strategy from the list set
elif self.mutation_type == 'fixed':
new_strategy = random.choice(
list(self.strategies_to_count.values()))
elif self.mutation_type == 'gaussian_sentimental':
# Copy the damn list
new_strategy = agent.strategy.copy()
# There is a 20% chance of mutation
if self.random.random() < 0.2:
# Each Pi is mutated by a value drawn from a gaussian
# with mean=delta_energy
for i in range(4):
mutation = self.random.normalvariate(
(agent.delta_energy + self.alpha()) / 14, self.d)
new_val = new_strategy[i] + mutation
# Keep probabilities in [0, 1]
new_val = (0 if new_val < 0 else
1 if new_val > 1 else new_val)
new_strategy[i] = new_val
return new_strategy
def maybe_reproduce(self, agent):
# If we have the energy to reproduce, do so
if agent.total_energy >= self.p:
# Create the child
new_strategy = self.maybe_mutate(agent)
child = GTAgent(self.agent_idx, agent.group_id, self, new_strategy,
self.i_energy)
self.agent_idx += 1
# Set parent and child energy levels to p/2
child.total_energy = self.p / 2
agent.total_energy = self.p / 2
# Place child (Remove agent argument for global child placement)
self.schedule.add(child)
self.grid.place_agent(child, self.get_child_location(agent))
def step(self):
if self.debug:
print('\n\n==================================================')
print('==================================================')
print('==================================================')
pprint(vars(self))
# First collect data
self.datacollector.collect(self)
# Then check for dead agents and for new agents
for agent in self.schedule.agent_buffer(shuffled=True):
# First check if dead
if self.time_to_die(agent):
self.grid.remove_agent(agent)
self.schedule.remove(agent)
# Otherwise check if can reproduce
else:
self.maybe_reproduce(agent)
# Finally, step each agent
self.schedule.step()
def check_strategy(self, agent):
# Helper function to check which strategy an agent would count as
def is_same(strategy, a_strategy):
tol = self.count_tolerance
return all(strategy[i] - tol < a_strategy[i] < strategy[i] + tol
for i in range(4))
return [
name for name, strat in self.strategies_to_count.items()
if is_same(strat, agent.strategy)
]