mirror of
https://github.com/gsi-upm/soil
synced 2024-11-22 11:12:29 +00:00
Separate models in modules
This commit is contained in:
parent
d157a5e0b5
commit
dd4ce15a3d
40
models/BaseBehaviour/BaseBehaviour.py
Normal file
40
models/BaseBehaviour/BaseBehaviour.py
Normal file
@ -0,0 +1,40 @@
|
|||||||
|
import settings
|
||||||
|
from nxsim import BaseNetworkAgent
|
||||||
|
from .. import networkStatus
|
||||||
|
|
||||||
|
settings.init()
|
||||||
|
|
||||||
|
|
||||||
|
class BaseBehaviour(BaseNetworkAgent):
|
||||||
|
|
||||||
|
def __init__(self, environment=None, agent_id=0, state=()):
|
||||||
|
super().__init__(environment=environment, agent_id=agent_id, state=state)
|
||||||
|
self._attrs = {}
|
||||||
|
|
||||||
|
@property
|
||||||
|
def attrs(self):
|
||||||
|
now = self.env.now
|
||||||
|
if now not in self._attrs:
|
||||||
|
self._attrs[now] = {}
|
||||||
|
return self._attrs[now]
|
||||||
|
|
||||||
|
@attrs.setter
|
||||||
|
def attrs(self, value):
|
||||||
|
self._attrs[self.env.now] = value
|
||||||
|
|
||||||
|
def run(self):
|
||||||
|
while True:
|
||||||
|
self.step(self.env.now)
|
||||||
|
yield self.env.timeout(settings.timeout)
|
||||||
|
|
||||||
|
def step(self, now):
|
||||||
|
networkStatus['agent_%s'% self.id] = self.to_json()
|
||||||
|
|
||||||
|
def to_json(self):
|
||||||
|
final = {}
|
||||||
|
for stamp, attrs in self._attrs.items():
|
||||||
|
for a in attrs:
|
||||||
|
if a not in final:
|
||||||
|
final[a] = {}
|
||||||
|
final[a][stamp] = attrs[a]
|
||||||
|
return final
|
1
models/BaseBehaviour/__init__.py
Normal file
1
models/BaseBehaviour/__init__.py
Normal file
@ -0,0 +1 @@
|
|||||||
|
from .BaseBehaviour import BaseBehaviour
|
49
models/BassModel/BassModel.py
Normal file
49
models/BassModel/BassModel.py
Normal file
@ -0,0 +1,49 @@
|
|||||||
|
import settings
|
||||||
|
import random
|
||||||
|
from ..BaseBehaviour import *
|
||||||
|
from .. import sentimentCorrelationNodeArray
|
||||||
|
|
||||||
|
settings.init()
|
||||||
|
|
||||||
|
|
||||||
|
class BassModel(BaseBehaviour):
|
||||||
|
"""
|
||||||
|
Settings:
|
||||||
|
innovation_prob
|
||||||
|
|
||||||
|
imitation_prob
|
||||||
|
"""
|
||||||
|
|
||||||
|
def __init__(self, environment=None, agent_id=0, state=()):
|
||||||
|
super().__init__(environment=environment, agent_id=agent_id, state=state)
|
||||||
|
self.innovation_prob = settings.innovation_prob
|
||||||
|
self.imitation_prob = settings.imitation_prob
|
||||||
|
sentimentCorrelationNodeArray[self.id][self.env.now]=0
|
||||||
|
|
||||||
|
def step(self, now):
|
||||||
|
self.behaviour()
|
||||||
|
super().step(now)
|
||||||
|
|
||||||
|
def behaviour(self):
|
||||||
|
# Outside effects
|
||||||
|
if random.random() < settings.innovation_prob:
|
||||||
|
if self.state['id'] == 0:
|
||||||
|
self.state['id'] = 1
|
||||||
|
sentimentCorrelationNodeArray[self.id][self.env.now]=1
|
||||||
|
else:
|
||||||
|
pass
|
||||||
|
|
||||||
|
self.attrs['status'] = self.state['id']
|
||||||
|
return
|
||||||
|
|
||||||
|
# Imitation effects
|
||||||
|
if self.state['id'] == 0:
|
||||||
|
aware_neighbors = self.get_neighboring_agents(state_id=1)
|
||||||
|
num_neighbors_aware = len(aware_neighbors)
|
||||||
|
if random.random() < (settings.imitation_prob*num_neighbors_aware):
|
||||||
|
self.state['id'] = 1
|
||||||
|
sentimentCorrelationNodeArray[self.id][self.env.now]=1
|
||||||
|
|
||||||
|
else:
|
||||||
|
pass
|
||||||
|
self.attrs['status'] = self.state['id']
|
1
models/BassModel/__init__.py
Normal file
1
models/BassModel/__init__.py
Normal file
@ -0,0 +1 @@
|
|||||||
|
from .BassModel import BassModel
|
108
models/BigMarketModel/BigMarketModel.py
Normal file
108
models/BigMarketModel/BigMarketModel.py
Normal file
@ -0,0 +1,108 @@
|
|||||||
|
import settings
|
||||||
|
import random
|
||||||
|
from ..BaseBehaviour import *
|
||||||
|
|
||||||
|
settings.init()
|
||||||
|
|
||||||
|
|
||||||
|
class BigMarketModel(BaseBehaviour):
|
||||||
|
"""
|
||||||
|
Settings:
|
||||||
|
Names:
|
||||||
|
enterprises [Array]
|
||||||
|
|
||||||
|
tweet_probability_enterprises [Array]
|
||||||
|
Users:
|
||||||
|
tweet_probability_users
|
||||||
|
|
||||||
|
tweet_relevant_probability
|
||||||
|
|
||||||
|
tweet_probability_about [Array]
|
||||||
|
|
||||||
|
sentiment_about [Array]
|
||||||
|
"""
|
||||||
|
|
||||||
|
|
||||||
|
def __init__(self, environment=None, agent_id=0, state=()):
|
||||||
|
super().__init__(environment=environment, agent_id=agent_id, state=state)
|
||||||
|
self.enterprises = settings.enterprises
|
||||||
|
self.type = ""
|
||||||
|
self.number_of_enterprises = len(settings.enterprises)
|
||||||
|
|
||||||
|
if self.id < self.number_of_enterprises: # Enterprises
|
||||||
|
self.state['id']=self.id
|
||||||
|
self.type="Enterprise"
|
||||||
|
self.tweet_probability = settings.tweet_probability_enterprises[self.id]
|
||||||
|
else: # normal users
|
||||||
|
self.state['id']=self.number_of_enterprises
|
||||||
|
self.type="User"
|
||||||
|
self.tweet_probability = settings.tweet_probability_users
|
||||||
|
self.tweet_relevant_probability = settings.tweet_relevant_probability
|
||||||
|
self.tweet_probability_about = settings.tweet_probability_about # List
|
||||||
|
self.sentiment_about = settings.sentiment_about # List
|
||||||
|
|
||||||
|
def step(self, now):
|
||||||
|
|
||||||
|
if(self.id < self.number_of_enterprises): # Enterprise
|
||||||
|
self.enterpriseBehaviour()
|
||||||
|
else: # Usuario
|
||||||
|
self.userBehaviour()
|
||||||
|
for i in range(self.number_of_enterprises): # So that it never is set to 0 if there are not changes (logs)
|
||||||
|
self.attrs['sentiment_enterprise_%s'% self.enterprises[i]] = self.sentiment_about[i]
|
||||||
|
|
||||||
|
super().step(now)
|
||||||
|
|
||||||
|
def enterpriseBehaviour(self):
|
||||||
|
|
||||||
|
if random.random()< self.tweet_probability: # Tweets
|
||||||
|
aware_neighbors = self.get_neighboring_agents(state_id=self.number_of_enterprises) #Nodes neighbour users
|
||||||
|
for x in aware_neighbors:
|
||||||
|
if random.uniform(0,10) < 5:
|
||||||
|
x.sentiment_about[self.id] += 0.1 # Increments for enterprise
|
||||||
|
else:
|
||||||
|
x.sentiment_about[self.id] -= 0.1 # Decrements for enterprise
|
||||||
|
|
||||||
|
# Establecemos limites
|
||||||
|
if x.sentiment_about[self.id] > 1:
|
||||||
|
x.sentiment_about[self.id] = 1
|
||||||
|
if x.sentiment_about[self.id]< -1:
|
||||||
|
x.sentiment_about[self.id] = -1
|
||||||
|
|
||||||
|
x.attrs['sentiment_enterprise_%s'% self.enterprises[self.id]] = x.sentiment_about[self.id]
|
||||||
|
|
||||||
|
def userBehaviour(self):
|
||||||
|
|
||||||
|
if random.random() < self.tweet_probability: # Tweets
|
||||||
|
if random.random() < self.tweet_relevant_probability: # Tweets something relevant
|
||||||
|
# Tweet probability per enterprise
|
||||||
|
for i in range(self.number_of_enterprises):
|
||||||
|
random_num = random.random()
|
||||||
|
if random_num < self.tweet_probability_about[i]:
|
||||||
|
# The condition is fulfilled, sentiments are evaluated towards that enterprise
|
||||||
|
if self.sentiment_about[i] < 0:
|
||||||
|
# NEGATIVO
|
||||||
|
self.userTweets("negative",i)
|
||||||
|
elif self.sentiment_about[i] == 0:
|
||||||
|
# NEUTRO
|
||||||
|
pass
|
||||||
|
else:
|
||||||
|
# POSITIVO
|
||||||
|
self.userTweets("positive",i)
|
||||||
|
|
||||||
|
def userTweets(self,sentiment,enterprise):
|
||||||
|
aware_neighbors = self.get_neighboring_agents(state_id=self.number_of_enterprises) #Nodes neighbours users
|
||||||
|
for x in aware_neighbors:
|
||||||
|
if sentiment == "positive":
|
||||||
|
x.sentiment_about[enterprise] +=0.003
|
||||||
|
elif sentiment == "negative":
|
||||||
|
x.sentiment_about[enterprise] -=0.003
|
||||||
|
else:
|
||||||
|
pass
|
||||||
|
|
||||||
|
# Establecemos limites
|
||||||
|
if x.sentiment_about[enterprise] > 1:
|
||||||
|
x.sentiment_about[enterprise] = 1
|
||||||
|
if x.sentiment_about[enterprise] < -1:
|
||||||
|
x.sentiment_about[enterprise] = -1
|
||||||
|
|
||||||
|
x.attrs['sentiment_enterprise_%s'% self.enterprises[enterprise]] = x.sentiment_about[enterprise]
|
96
models/BigMarketModel/SISaModel.py
Normal file
96
models/BigMarketModel/SISaModel.py
Normal file
@ -0,0 +1,96 @@
|
|||||||
|
import settings
|
||||||
|
import random
|
||||||
|
import numpy as np
|
||||||
|
from ..BaseBehaviour import *
|
||||||
|
|
||||||
|
settings.init()
|
||||||
|
|
||||||
|
|
||||||
|
class SISaModel(BaseBehaviour):
|
||||||
|
"""
|
||||||
|
Settings:
|
||||||
|
neutral_discontent_spon_prob
|
||||||
|
|
||||||
|
neutral_discontent_infected_prob
|
||||||
|
|
||||||
|
neutral_content_spong_prob
|
||||||
|
|
||||||
|
neutral_content_infected_prob
|
||||||
|
|
||||||
|
discontent_neutral
|
||||||
|
|
||||||
|
discontent_content
|
||||||
|
|
||||||
|
variance_d_c
|
||||||
|
|
||||||
|
content_discontent
|
||||||
|
|
||||||
|
variance_c_d
|
||||||
|
|
||||||
|
content_neutral
|
||||||
|
|
||||||
|
standard_variance
|
||||||
|
"""
|
||||||
|
|
||||||
|
def __init__(self, environment=None, agent_id=0, state=()):
|
||||||
|
super().__init__(environment=environment, agent_id=agent_id, state=state)
|
||||||
|
|
||||||
|
self.neutral_discontent_spon_prob = np.random.normal(settings.neutral_discontent_spon_prob,
|
||||||
|
settings.standard_variance)
|
||||||
|
self.neutral_discontent_infected_prob = np.random.normal(settings.neutral_discontent_infected_prob,
|
||||||
|
settings.standard_variance)
|
||||||
|
self.neutral_content_spon_prob = np.random.normal(settings.neutral_content_spon_prob, settings.standard_variance)
|
||||||
|
self.neutral_content_infected_prob = np.random.normal(settings.neutral_content_infected_prob,
|
||||||
|
settings.standard_variance)
|
||||||
|
|
||||||
|
self.discontent_neutral = np.random.normal(settings.discontent_neutral, settings.standard_variance)
|
||||||
|
self.discontent_content = np.random.normal(settings.discontent_content, settings.variance_d_c)
|
||||||
|
|
||||||
|
self.content_discontent = np.random.normal(settings.content_discontent, settings.variance_c_d)
|
||||||
|
self.content_neutral = np.random.normal(settings.content_neutral, settings.standard_variance)
|
||||||
|
|
||||||
|
def step(self, now):
|
||||||
|
if self.state['id'] == 0:
|
||||||
|
self.neutral_behaviour()
|
||||||
|
if self.state['id'] == 1:
|
||||||
|
self.discontent_behaviour()
|
||||||
|
if self.state['id'] == 2:
|
||||||
|
self.content_behaviour()
|
||||||
|
|
||||||
|
self.attrs['status'] = self.state['id']
|
||||||
|
super().step(now)
|
||||||
|
|
||||||
|
def neutral_behaviour(self):
|
||||||
|
# Spontaneous effects
|
||||||
|
if random.random() < self.neutral_discontent_spon_prob:
|
||||||
|
self.state['id'] = 1
|
||||||
|
if random.random() < self.neutral_content_spon_prob:
|
||||||
|
self.state['id'] = 2
|
||||||
|
|
||||||
|
# Infected
|
||||||
|
discontent_neighbors = self.get_neighboring_agents(state_id=1)
|
||||||
|
if random.random() < len(discontent_neighbors) * self.neutral_discontent_infected_prob:
|
||||||
|
self.state['id'] = 1
|
||||||
|
content_neighbors = self.get_neighboring_agents(state_id=2)
|
||||||
|
if random.random() < len(content_neighbors) * self.neutral_content_infected_prob:
|
||||||
|
self.state['id'] = 2
|
||||||
|
|
||||||
|
def discontent_behaviour(self):
|
||||||
|
# Healing
|
||||||
|
if random.random() < self.discontent_neutral:
|
||||||
|
self.state['id'] = 0
|
||||||
|
|
||||||
|
# Superinfected
|
||||||
|
content_neighbors = self.get_neighboring_agents(state_id=2)
|
||||||
|
if random.random() < len(content_neighbors) * self.discontent_content:
|
||||||
|
self.state['id'] = 2
|
||||||
|
|
||||||
|
def content_behaviour(self):
|
||||||
|
# Healing
|
||||||
|
if random.random() < self.content_neutral:
|
||||||
|
self.state['id'] = 0
|
||||||
|
|
||||||
|
# Superinfected
|
||||||
|
discontent_neighbors = self.get_neighboring_agents(state_id=1)
|
||||||
|
if random.random() < len(discontent_neighbors) * self.content_discontent:
|
||||||
|
self.state['id'] = 1
|
2
models/BigMarketModel/__init__.py
Normal file
2
models/BigMarketModel/__init__.py
Normal file
@ -0,0 +1,2 @@
|
|||||||
|
from .BigMarketModel import BigMarketModel
|
||||||
|
from .SISaModel import SISaModel
|
56
models/IndependentCascadeModel/IndependentCascadeModel.py
Normal file
56
models/IndependentCascadeModel/IndependentCascadeModel.py
Normal file
@ -0,0 +1,56 @@
|
|||||||
|
import settings
|
||||||
|
import random
|
||||||
|
from ..BaseBehaviour import *
|
||||||
|
from .. import sentimentCorrelationNodeArray
|
||||||
|
|
||||||
|
settings.init()
|
||||||
|
|
||||||
|
|
||||||
|
class IndependentCascadeModel(BaseBehaviour):
|
||||||
|
"""
|
||||||
|
Settings:
|
||||||
|
innovation_prob
|
||||||
|
|
||||||
|
imitation_prob
|
||||||
|
"""
|
||||||
|
|
||||||
|
def __init__(self, environment=None, agent_id=0, state=()):
|
||||||
|
super().__init__(environment=environment, agent_id=agent_id, state=state)
|
||||||
|
self.innovation_prob = settings.innovation_prob
|
||||||
|
self.imitation_prob = settings.imitation_prob
|
||||||
|
self.time_awareness = 0
|
||||||
|
sentimentCorrelationNodeArray[self.id][self.env.now]=0
|
||||||
|
|
||||||
|
def step(self,now):
|
||||||
|
self.behaviour()
|
||||||
|
super().step(now)
|
||||||
|
|
||||||
|
def behaviour(self):
|
||||||
|
aware_neighbors_1_time_step=[]
|
||||||
|
# Outside effects
|
||||||
|
if random.random() < settings.innovation_prob:
|
||||||
|
if self.state['id'] == 0:
|
||||||
|
self.state['id'] = 1
|
||||||
|
sentimentCorrelationNodeArray[self.id][self.env.now]=1
|
||||||
|
self.time_awareness = self.env.now #To know when they have been infected
|
||||||
|
else:
|
||||||
|
pass
|
||||||
|
|
||||||
|
self.attrs['status'] = self.state['id']
|
||||||
|
return
|
||||||
|
|
||||||
|
# Imitation effects
|
||||||
|
if self.state['id'] == 0:
|
||||||
|
aware_neighbors = self.get_neighboring_agents(state_id=1)
|
||||||
|
for x in aware_neighbors:
|
||||||
|
if x.time_awareness == (self.env.now-1):
|
||||||
|
aware_neighbors_1_time_step.append(x)
|
||||||
|
num_neighbors_aware = len(aware_neighbors_1_time_step)
|
||||||
|
if random.random() < (settings.imitation_prob*num_neighbors_aware):
|
||||||
|
self.state['id'] = 1
|
||||||
|
sentimentCorrelationNodeArray[self.id][self.env.now]=1
|
||||||
|
else:
|
||||||
|
pass
|
||||||
|
|
||||||
|
self.attrs['status'] = self.state['id']
|
||||||
|
return
|
1
models/IndependentCascadeModel/__init__.py
Normal file
1
models/IndependentCascadeModel/__init__.py
Normal file
@ -0,0 +1 @@
|
|||||||
|
from .IndependentCascadeModel import IndependentCascadeModel
|
143
models/ModelM2/ControlModelM2.py
Normal file
143
models/ModelM2/ControlModelM2.py
Normal file
@ -0,0 +1,143 @@
|
|||||||
|
import settings
|
||||||
|
import random
|
||||||
|
import numpy as np
|
||||||
|
from ..BaseBehaviour import *
|
||||||
|
from .. import init_states
|
||||||
|
|
||||||
|
settings.init()
|
||||||
|
|
||||||
|
|
||||||
|
class ControlModelM2(BaseBehaviour):
|
||||||
|
"""
|
||||||
|
Settings:
|
||||||
|
prob_neutral_making_denier
|
||||||
|
|
||||||
|
prob_infect
|
||||||
|
|
||||||
|
prob_cured_healing_infected
|
||||||
|
|
||||||
|
prob_cured_vaccinate_neutral
|
||||||
|
|
||||||
|
prob_vaccinated_healing_infected
|
||||||
|
|
||||||
|
prob_vaccinated_vaccinate_neutral
|
||||||
|
|
||||||
|
prob_generate_anti_rumor
|
||||||
|
"""
|
||||||
|
|
||||||
|
# Init infected
|
||||||
|
init_states[random.randint(0, settings.number_of_nodes-1)] = {'id':1}
|
||||||
|
init_states[random.randint(0, settings.number_of_nodes-1)] = {'id':1}
|
||||||
|
|
||||||
|
# Init beacons
|
||||||
|
init_states[random.randint(0, settings.number_of_nodes-1)] = {'id': 4}
|
||||||
|
init_states[random.randint(0, settings.number_of_nodes-1)] = {'id': 4}
|
||||||
|
|
||||||
|
def __init__(self, environment=None, agent_id=0, state=()):
|
||||||
|
super().__init__(environment=environment, agent_id=agent_id, state=state)
|
||||||
|
|
||||||
|
self.prob_neutral_making_denier = np.random.normal(settings.prob_neutral_making_denier, settings.standard_variance)
|
||||||
|
|
||||||
|
self.prob_infect = np.random.normal(settings.prob_infect, settings.standard_variance)
|
||||||
|
|
||||||
|
self.prob_cured_healing_infected = np.random.normal(settings.prob_cured_healing_infected, settings.standard_variance)
|
||||||
|
self.prob_cured_vaccinate_neutral = np.random.normal(settings.prob_cured_vaccinate_neutral, settings.standard_variance)
|
||||||
|
|
||||||
|
self.prob_vaccinated_healing_infected = np.random.normal(settings.prob_vaccinated_healing_infected, settings.standard_variance)
|
||||||
|
self.prob_vaccinated_vaccinate_neutral = np.random.normal(settings.prob_vaccinated_vaccinate_neutral, settings.standard_variance)
|
||||||
|
self.prob_generate_anti_rumor = np.random.normal(settings.prob_generate_anti_rumor, settings.standard_variance)
|
||||||
|
|
||||||
|
def step(self, now):
|
||||||
|
|
||||||
|
if self.state['id'] == 0: # Neutral
|
||||||
|
self.neutral_behaviour()
|
||||||
|
elif self.state['id'] == 1: # Infected
|
||||||
|
self.infected_behaviour()
|
||||||
|
elif self.state['id'] == 2: # Cured
|
||||||
|
self.cured_behaviour()
|
||||||
|
elif self.state['id'] == 3: # Vaccinated
|
||||||
|
self.vaccinated_behaviour()
|
||||||
|
elif self.state['id'] == 4: # Beacon-off
|
||||||
|
self.beacon_off_behaviour()
|
||||||
|
elif self.state['id'] == 5: # Beacon-on
|
||||||
|
self.beacon_on_behaviour()
|
||||||
|
|
||||||
|
self.attrs['status'] = self.state['id']
|
||||||
|
super().step(now)
|
||||||
|
|
||||||
|
def neutral_behaviour(self):
|
||||||
|
|
||||||
|
# Infected
|
||||||
|
infected_neighbors = self.get_neighboring_agents(state_id=1)
|
||||||
|
if len(infected_neighbors)>0:
|
||||||
|
if random.random() < self.prob_neutral_making_denier:
|
||||||
|
self.state['id'] = 3 # Vaccinated making denier
|
||||||
|
|
||||||
|
def infected_behaviour(self):
|
||||||
|
|
||||||
|
# Neutral
|
||||||
|
neutral_neighbors = self.get_neighboring_agents(state_id=0)
|
||||||
|
for neighbor in neutral_neighbors:
|
||||||
|
if random.random() < self.prob_infect:
|
||||||
|
neighbor.state['id'] = 1 # Infected
|
||||||
|
|
||||||
|
def cured_behaviour(self):
|
||||||
|
|
||||||
|
# Vaccinate
|
||||||
|
neutral_neighbors = self.get_neighboring_agents(state_id=0)
|
||||||
|
for neighbor in neutral_neighbors:
|
||||||
|
if random.random() < self.prob_cured_vaccinate_neutral:
|
||||||
|
neighbor.state['id'] = 3 # Vaccinated
|
||||||
|
|
||||||
|
# Cure
|
||||||
|
infected_neighbors = self.get_neighboring_agents(state_id=1)
|
||||||
|
for neighbor in infected_neighbors:
|
||||||
|
if random.random() < self.prob_cured_healing_infected:
|
||||||
|
neighbor.state['id'] = 2 # Cured
|
||||||
|
|
||||||
|
def vaccinated_behaviour(self):
|
||||||
|
|
||||||
|
# Cure
|
||||||
|
infected_neighbors = self.get_neighboring_agents(state_id=1)
|
||||||
|
for neighbor in infected_neighbors:
|
||||||
|
if random.random() < self.prob_cured_healing_infected:
|
||||||
|
neighbor.state['id'] = 2 # Cured
|
||||||
|
|
||||||
|
# Vaccinate
|
||||||
|
neutral_neighbors = self.get_neighboring_agents(state_id=0)
|
||||||
|
for neighbor in neutral_neighbors:
|
||||||
|
if random.random() < self.prob_cured_vaccinate_neutral:
|
||||||
|
neighbor.state['id'] = 3 # Vaccinated
|
||||||
|
|
||||||
|
# Generate anti-rumor
|
||||||
|
infected_neighbors_2 = self.get_neighboring_agents(state_id=1)
|
||||||
|
for neighbor in infected_neighbors_2:
|
||||||
|
if random.random() < self.prob_generate_anti_rumor:
|
||||||
|
neighbor.state['id'] = 2 # Cured
|
||||||
|
|
||||||
|
def beacon_off_behaviour(self):
|
||||||
|
infected_neighbors = self.get_neighboring_agents(state_id=1)
|
||||||
|
if len(infected_neighbors) > 0:
|
||||||
|
self.state['id'] == 5 # Beacon on
|
||||||
|
|
||||||
|
def beacon_on_behaviour(self):
|
||||||
|
|
||||||
|
# Cure (M2 feature added)
|
||||||
|
infected_neighbors = self.get_neighboring_agents(state_id=1)
|
||||||
|
for neighbor in infected_neighbors:
|
||||||
|
if random.random() < self.prob_generate_anti_rumor:
|
||||||
|
neighbor.state['id'] = 2 # Cured
|
||||||
|
neutral_neighbors_infected = neighbor.get_neighboring_agents(state_id=0)
|
||||||
|
for neighbor in neutral_neighbors_infected:
|
||||||
|
if random.random() < self.prob_generate_anti_rumor:
|
||||||
|
neighbor.state['id'] = 3 # Vaccinated
|
||||||
|
infected_neighbors_infected = neighbor.get_neighboring_agents(state_id=1)
|
||||||
|
for neighbor in infected_neighbors_infected:
|
||||||
|
if random.random() < self.prob_generate_anti_rumor:
|
||||||
|
neighbor.state['id'] = 2 # Cured
|
||||||
|
|
||||||
|
# Vaccinate
|
||||||
|
neutral_neighbors = self.get_neighboring_agents(state_id=0)
|
||||||
|
for neighbor in neutral_neighbors:
|
||||||
|
if random.random() < self.prob_cured_vaccinate_neutral:
|
||||||
|
neighbor.state['id'] = 3 # Vaccinated
|
107
models/ModelM2/SpreadModelM2.py
Normal file
107
models/ModelM2/SpreadModelM2.py
Normal file
@ -0,0 +1,107 @@
|
|||||||
|
import settings
|
||||||
|
import random
|
||||||
|
import numpy as np
|
||||||
|
from ..BaseBehaviour import *
|
||||||
|
from .. import init_states
|
||||||
|
|
||||||
|
settings.init()
|
||||||
|
|
||||||
|
|
||||||
|
class SpreadModelM2(BaseBehaviour):
|
||||||
|
"""
|
||||||
|
Settings:
|
||||||
|
prob_neutral_making_denier
|
||||||
|
|
||||||
|
prob_infect
|
||||||
|
|
||||||
|
prob_cured_healing_infected
|
||||||
|
|
||||||
|
prob_cured_vaccinate_neutral
|
||||||
|
|
||||||
|
prob_vaccinated_healing_infected
|
||||||
|
|
||||||
|
prob_vaccinated_vaccinate_neutral
|
||||||
|
|
||||||
|
prob_generate_anti_rumor
|
||||||
|
"""
|
||||||
|
|
||||||
|
init_states[random.randint(0, settings.number_of_nodes)] = {'id':1}
|
||||||
|
init_states[random.randint(0, settings.number_of_nodes)] = {'id':1}
|
||||||
|
|
||||||
|
def __init__(self, environment=None, agent_id=0, state=()):
|
||||||
|
super().__init__(environment=environment, agent_id=agent_id, state=state)
|
||||||
|
|
||||||
|
self.prob_neutral_making_denier = np.random.normal(settings.prob_neutral_making_denier, settings.standard_variance)
|
||||||
|
|
||||||
|
self.prob_infect = np.random.normal(settings.prob_infect, settings.standard_variance)
|
||||||
|
|
||||||
|
self.prob_cured_healing_infected = np.random.normal(settings.prob_cured_healing_infected, settings.standard_variance)
|
||||||
|
self.prob_cured_vaccinate_neutral = np.random.normal(settings.prob_cured_vaccinate_neutral, settings.standard_variance)
|
||||||
|
|
||||||
|
self.prob_vaccinated_healing_infected = np.random.normal(settings.prob_vaccinated_healing_infected, settings.standard_variance)
|
||||||
|
self.prob_vaccinated_vaccinate_neutral = np.random.normal(settings.prob_vaccinated_vaccinate_neutral, settings.standard_variance)
|
||||||
|
self.prob_generate_anti_rumor = np.random.normal(settings.prob_generate_anti_rumor, settings.standard_variance)
|
||||||
|
|
||||||
|
def step(self, now):
|
||||||
|
|
||||||
|
if self.state['id'] == 0: # Neutral
|
||||||
|
self.neutral_behaviour()
|
||||||
|
elif self.state['id'] == 1: # Infected
|
||||||
|
self.infected_behaviour()
|
||||||
|
elif self.state['id'] == 2: # Cured
|
||||||
|
self.cured_behaviour()
|
||||||
|
elif self.state['id'] == 3: # Vaccinated
|
||||||
|
self.vaccinated_behaviour()
|
||||||
|
|
||||||
|
self.attrs['status'] = self.state['id']
|
||||||
|
super().step(now)
|
||||||
|
|
||||||
|
def neutral_behaviour(self):
|
||||||
|
|
||||||
|
# Infected
|
||||||
|
infected_neighbors = self.get_neighboring_agents(state_id=1)
|
||||||
|
if len(infected_neighbors)>0:
|
||||||
|
if random.random() < self.prob_neutral_making_denier:
|
||||||
|
self.state['id'] = 3 # Vaccinated making denier
|
||||||
|
|
||||||
|
def infected_behaviour(self):
|
||||||
|
|
||||||
|
# Neutral
|
||||||
|
neutral_neighbors = self.get_neighboring_agents(state_id=0)
|
||||||
|
for neighbor in neutral_neighbors:
|
||||||
|
if random.random() < self.prob_infect:
|
||||||
|
neighbor.state['id'] = 1 # Infected
|
||||||
|
|
||||||
|
def cured_behaviour(self):
|
||||||
|
|
||||||
|
# Vaccinate
|
||||||
|
neutral_neighbors = self.get_neighboring_agents(state_id=0)
|
||||||
|
for neighbor in neutral_neighbors:
|
||||||
|
if random.random() < self.prob_cured_vaccinate_neutral:
|
||||||
|
neighbor.state['id'] = 3 # Vaccinated
|
||||||
|
|
||||||
|
# Cure
|
||||||
|
infected_neighbors = self.get_neighboring_agents(state_id=1)
|
||||||
|
for neighbor in infected_neighbors:
|
||||||
|
if random.random() < self.prob_cured_healing_infected:
|
||||||
|
neighbor.state['id'] = 2 # Cured
|
||||||
|
|
||||||
|
def vaccinated_behaviour(self):
|
||||||
|
|
||||||
|
# Cure
|
||||||
|
infected_neighbors = self.get_neighboring_agents(state_id=1)
|
||||||
|
for neighbor in infected_neighbors:
|
||||||
|
if random.random() < self.prob_cured_healing_infected:
|
||||||
|
neighbor.state['id'] = 2 # Cured
|
||||||
|
|
||||||
|
# Vaccinate
|
||||||
|
neutral_neighbors = self.get_neighboring_agents(state_id=0)
|
||||||
|
for neighbor in neutral_neighbors:
|
||||||
|
if random.random() < self.prob_cured_vaccinate_neutral:
|
||||||
|
neighbor.state['id'] = 3 # Vaccinated
|
||||||
|
|
||||||
|
# Generate anti-rumor
|
||||||
|
infected_neighbors_2 = self.get_neighboring_agents(state_id=1)
|
||||||
|
for neighbor in infected_neighbors_2:
|
||||||
|
if random.random() < self.prob_generate_anti_rumor:
|
||||||
|
neighbor.state['id'] = 2 # Cured
|
2
models/ModelM2/__init__.py
Normal file
2
models/ModelM2/__init__.py
Normal file
@ -0,0 +1,2 @@
|
|||||||
|
from .ControlModelM2 import ControlModelM2
|
||||||
|
from .SpreadModelM2 import SpreadModelM2
|
109
models/SentimentCorrelationModel/SentimentCorrelationModel.py
Normal file
109
models/SentimentCorrelationModel/SentimentCorrelationModel.py
Normal file
@ -0,0 +1,109 @@
|
|||||||
|
import settings
|
||||||
|
import random
|
||||||
|
from ..BaseBehaviour import *
|
||||||
|
from .. import sentimentCorrelationNodeArray
|
||||||
|
|
||||||
|
settings.init()
|
||||||
|
|
||||||
|
|
||||||
|
class SentimentCorrelationModel(BaseBehaviour):
|
||||||
|
"""
|
||||||
|
Settings:
|
||||||
|
outside_effects_prob
|
||||||
|
|
||||||
|
anger_prob
|
||||||
|
|
||||||
|
joy_prob
|
||||||
|
|
||||||
|
sadness_prob
|
||||||
|
|
||||||
|
disgust_prob
|
||||||
|
"""
|
||||||
|
|
||||||
|
def __init__(self, environment=None, agent_id=0, state=()):
|
||||||
|
super().__init__(environment=environment, agent_id=agent_id, state=state)
|
||||||
|
self.outside_effects_prob = settings.outside_effects_prob
|
||||||
|
self.anger_prob = settings.anger_prob
|
||||||
|
self.joy_prob = settings.joy_prob
|
||||||
|
self.sadness_prob = settings.sadness_prob
|
||||||
|
self.disgust_prob = settings.disgust_prob
|
||||||
|
self.time_awareness=[]
|
||||||
|
for i in range(4): #In this model we have 4 sentiments
|
||||||
|
self.time_awareness.append(0) #0-> Anger, 1-> joy, 2->sadness, 3 -> disgust
|
||||||
|
sentimentCorrelationNodeArray[self.id][self.env.now]=0
|
||||||
|
|
||||||
|
def step(self, now):
|
||||||
|
self.behaviour()
|
||||||
|
super().step(now)
|
||||||
|
|
||||||
|
def behaviour(self):
|
||||||
|
|
||||||
|
angry_neighbors_1_time_step=[]
|
||||||
|
joyful_neighbors_1_time_step=[]
|
||||||
|
sad_neighbors_1_time_step=[]
|
||||||
|
disgusted_neighbors_1_time_step=[]
|
||||||
|
|
||||||
|
angry_neighbors = self.get_neighboring_agents(state_id=1)
|
||||||
|
for x in angry_neighbors:
|
||||||
|
if x.time_awareness[0] > (self.env.now-500):
|
||||||
|
angry_neighbors_1_time_step.append(x)
|
||||||
|
num_neighbors_angry = len(angry_neighbors_1_time_step)
|
||||||
|
|
||||||
|
joyful_neighbors = self.get_neighboring_agents(state_id=2)
|
||||||
|
for x in joyful_neighbors:
|
||||||
|
if x.time_awareness[1] > (self.env.now-500):
|
||||||
|
joyful_neighbors_1_time_step.append(x)
|
||||||
|
num_neighbors_joyful = len(joyful_neighbors_1_time_step)
|
||||||
|
|
||||||
|
sad_neighbors = self.get_neighboring_agents(state_id=3)
|
||||||
|
for x in sad_neighbors:
|
||||||
|
if x.time_awareness[2] > (self.env.now-500):
|
||||||
|
sad_neighbors_1_time_step.append(x)
|
||||||
|
num_neighbors_sad = len(sad_neighbors_1_time_step)
|
||||||
|
|
||||||
|
disgusted_neighbors = self.get_neighboring_agents(state_id=4)
|
||||||
|
for x in disgusted_neighbors:
|
||||||
|
if x.time_awareness[3] > (self.env.now-500):
|
||||||
|
disgusted_neighbors_1_time_step.append(x)
|
||||||
|
num_neighbors_disgusted = len(disgusted_neighbors_1_time_step)
|
||||||
|
|
||||||
|
anger_prob= settings.anger_prob+(len(angry_neighbors_1_time_step)*settings.anger_prob)
|
||||||
|
joy_prob= settings.joy_prob+(len(joyful_neighbors_1_time_step)*settings.joy_prob)
|
||||||
|
sadness_prob = settings.sadness_prob+(len(sad_neighbors_1_time_step)*settings.sadness_prob)
|
||||||
|
disgust_prob = settings.disgust_prob+(len(disgusted_neighbors_1_time_step)*settings.disgust_prob)
|
||||||
|
outside_effects_prob= settings.outside_effects_prob
|
||||||
|
|
||||||
|
num = random.random()
|
||||||
|
|
||||||
|
if(num<outside_effects_prob):
|
||||||
|
self.state['id'] = random.randint(1,4)
|
||||||
|
|
||||||
|
sentimentCorrelationNodeArray[self.id][self.env.now]=self.state['id'] #It is stored when it has been infected for the dynamic network
|
||||||
|
self.time_awareness[self.state['id']-1] = self.env.now
|
||||||
|
self.attrs['sentiment'] = self.state['id']
|
||||||
|
|
||||||
|
|
||||||
|
if(num<anger_prob):
|
||||||
|
|
||||||
|
self.state['id'] = 1
|
||||||
|
sentimentCorrelationNodeArray[self.id][self.env.now]=1
|
||||||
|
self.time_awareness[self.state['id']-1] = self.env.now
|
||||||
|
elif (num<joy_prob+anger_prob and num>anger_prob):
|
||||||
|
|
||||||
|
self.state['id'] = 2
|
||||||
|
sentimentCorrelationNodeArray[self.id][self.env.now]=2
|
||||||
|
self.time_awareness[self.state['id']-1] = self.env.now
|
||||||
|
elif (num<sadness_prob+anger_prob+joy_prob and num>joy_prob+anger_prob):
|
||||||
|
|
||||||
|
|
||||||
|
self.state['id'] = 3
|
||||||
|
sentimentCorrelationNodeArray[self.id][self.env.now]=3
|
||||||
|
self.time_awareness[self.state['id']-1] = self.env.now
|
||||||
|
elif (num<disgust_prob+sadness_prob+anger_prob+joy_prob and num>sadness_prob+anger_prob+joy_prob):
|
||||||
|
|
||||||
|
|
||||||
|
self.state['id'] = 4
|
||||||
|
sentimentCorrelationNodeArray[self.id][self.env.now]=4
|
||||||
|
self.time_awareness[self.state['id']-1] = self.env.now
|
||||||
|
|
||||||
|
self.attrs['sentiment'] = self.state['id']
|
1
models/SentimentCorrelationModel/__init__.py
Normal file
1
models/SentimentCorrelationModel/__init__.py
Normal file
@ -0,0 +1 @@
|
|||||||
|
from .SentimentCorrelationModel import SentimentCorrelationModel
|
8
models/__init__.py
Normal file
8
models/__init__.py
Normal file
@ -0,0 +1,8 @@
|
|||||||
|
from .models import *
|
||||||
|
from .BaseBehaviour import *
|
||||||
|
from .BassModel import *
|
||||||
|
from .BigMarketModel import *
|
||||||
|
from .IndependentCascadeModel import *
|
||||||
|
from .ModelM2 import *
|
||||||
|
from .SentimentCorrelationModel import *
|
||||||
|
|
13
models/models.py
Normal file
13
models/models.py
Normal file
@ -0,0 +1,13 @@
|
|||||||
|
import settings
|
||||||
|
|
||||||
|
settings.init()
|
||||||
|
|
||||||
|
global networkStatus
|
||||||
|
networkStatus = {} # Dict that will contain the status of every agent in the network
|
||||||
|
|
||||||
|
sentimentCorrelationNodeArray=[]
|
||||||
|
for x in range(0, settings.number_of_nodes):
|
||||||
|
sentimentCorrelationNodeArray.append({'id':x})
|
||||||
|
# Initialize agent states. Let's assume everyone is normal.
|
||||||
|
init_states = [{'id': 0, } for _ in range(settings.number_of_nodes)]
|
||||||
|
# add keys as as necessary, but "id" must always refer to that state category
|
1
soil.py
1
soil.py
@ -9,7 +9,6 @@ import math
|
|||||||
import json
|
import json
|
||||||
|
|
||||||
settings.init() # Loads all the data from settings
|
settings.init() # Loads all the data from settings
|
||||||
models.init() # Loads the models and network variables
|
|
||||||
|
|
||||||
####################
|
####################
|
||||||
# Network creation #
|
# Network creation #
|
||||||
|
Loading…
Reference in New Issue
Block a user