Coverage for .tox/p311/lib/python3.11/site-packages/scicom/historicalletters/model.py: 95%
110 statements
« prev ^ index » next coverage.py v7.4.4, created at 2024-05-15 13:48 +0200
« prev ^ index » next coverage.py v7.4.4, created at 2024-05-15 13:48 +0200
1"""The model class for HistoricalLetters."""
2import random
3from pathlib import Path
5import mesa
6import mesa_geo as mg
7import networkx as nx
8import pandas as pd
9from numpy import mean
10from shapely import contains
11from tqdm import tqdm
13from scicom.historicalletters.agents import RegionAgent, SenderAgent
14from scicom.historicalletters.space import Nuts2Eu
15from scicom.historicalletters.utils import createData
16from scicom.utilities.statistics import prune
19def getPrunedLedger(model: mesa.Model) -> pd.DataFrame:
20 """Model reporter for simulation of archiving.
22 Returns statistics of ledger network of model run
23 and various iterations of statistics of pruned networks.
25 The routine assumes that the network contains fields of sender,
26 receiver and step information.
27 """
28 if model.runPruning is True:
29 ledgerColumns = ["sender", "receiver", "sender_location", "receiver_location", "topic", "step"]
30 modelparams = {
31 "population": model.population,
32 "moveRange": model.moveRange,
33 "letterRange": model.letterRange,
34 "useActivation": model.useActivation,
35 "useSocialNetwork": model.useSocialNetwork,
36 "similarityThreshold": model.similarityThreshold,
37 "longRangeNetworkFactor": model.longRangeNetworkFactor,
38 "shortRangeNetworkFactor": model.shortRangeNetworkFactor,
39 }
40 result = prune(
41 modelparameters=modelparams,
42 network=model.letterLedger,
43 columns=ledgerColumns,
44 iterations=3,
45 delAmounts=(0.1, 0.25, 0.5, 0.75, 0.9),
46 delTypes=("unif", "exp", "beta", "log_normal1", "log_normal2", "log_normal3"),
47 delMethod=("agents", "regions", "time"),
48 rankedVals=(True, False),
49 )
50 else:
51 result = model.letterLedger
52 return result
55def getComponents(model: mesa.Model) -> int:
56 """Model reporter to get number of components.
58 The MultiDiGraph is converted to undirected,
59 considering only edges that are reciprocal, ie.
60 edges are established if sender and receiver have
61 exchanged at least a letter in each direction.
62 """
63 newg = model.socialNetwork.to_undirected(reciprocal=True)
64 return nx.number_connected_components(newg)
67def getScaledLetters(model: mesa.Model) -> float:
68 """Return relative number of send letters."""
69 return len(model.letterLedger)/model.schedule.time
72def getScaledMovements(model: mesa.Model) -> float:
73 """Return relative number of movements."""
74 return model.movements/model.schedule.time
77class HistoricalLetters(mesa.Model):
78 """A letter sending model with historical informed initital positions.
80 Each agent has an initial topic vector, expressed as a RGB value. The
81 initial positions of the agents is based on a weighted random draw
82 based on data from [1].
84 Each step, agents generate two neighbourhoods for sending letters and
85 potential targets to move towards. The probability to send letters is
86 a self-reinforcing process. During each sending the internal topic of
87 the sender is updated as a random rotation towards the receivers topic.
89 [1] J. Lobo et al, Population-Area Relationship for Medieval European Cities,
90 PLoS ONE 11(10): e0162678.
91 """
93 def __init__(
94 self,
95 population: int = 100,
96 moveRange: float = 0.05,
97 letterRange: float = 0.2,
98 similarityThreshold: float = 0.2,
99 longRangeNetworkFactor: float = 0.3,
100 shortRangeNetworkFactor: float = 0.4,
101 regionData: str = Path(Path(__file__).parent.parent.resolve(), "data/NUTS_RG_60M_2021_3857_LEVL_2.geojson"),
102 populationDistributionData: str = Path(Path(__file__).parent.parent.resolve(), "data/pone.0162678.s003.csv"),
103 *,
104 useActivation: bool = False,
105 useSocialNetwork: bool = False,
106 runPruning: bool = False,
107 debug: bool = False,
108 ) -> None:
109 """Initialize a HistoricalLetters model."""
110 super().__init__()
112 # Parameters for agents
113 self.population = population
114 self.moveRange = moveRange
115 self.letterRange = letterRange
116 # Parameters for model
117 self.runPruning = runPruning
118 self.useActivation = useActivation
119 self.similarityThreshold = similarityThreshold
120 self.useSocialNetwork = useSocialNetwork
121 self.longRangeNetworkFactor = longRangeNetworkFactor
122 self.shortRangeNetworkFactor = shortRangeNetworkFactor
123 # Initialize social network
124 self.socialNetwork = nx.MultiDiGraph()
125 # Output variables
126 self.letterLedger = []
127 self.movements = 0
128 # Internal variables
129 self.schedule = mesa.time.RandomActivation(self)
130 self.scaleSendInput = {}
131 self.updatedTopicsDict = {}
132 self.space = Nuts2Eu()
133 self.debug = debug
135 #######
136 # Initialize region agents
137 #######
139 # Set up the grid with patches for every NUTS region
140 # Create region agents
141 ac = mg.AgentCreator(RegionAgent, model=self)
142 self.regions = ac.from_file(
143 regionData,
144 unique_id="NUTS_ID",
145 )
146 # Add regions to Nuts2Eu geospace
147 self.space.add_regions(self.regions)
149 #######
150 # Initialize sender agents
151 #######
153 # Draw initial geographic positions of agents
154 initSenderGeoDf = createData(
155 population,
156 populationDistribution=populationDistributionData,
157 )
159 # Calculate mean of mean distances for each agent.
160 # This is used as a measure for the range of exchanges.
161 meandistances = []
162 for idx in initSenderGeoDf.index.to_numpy():
163 name = initSenderGeoDf.loc[idx, "unique_id"]
164 geom = initSenderGeoDf.loc[idx, "geometry"]
165 otherAgents = initSenderGeoDf.query(f"unique_id != '{name}'").copy()
166 geometries = otherAgents.geometry.to_numpy()
167 distances = [geom.distance(othergeom) for othergeom in geometries]
168 meandistances.append(mean(distances))
169 self.meandistance = mean(meandistances)
171 # Populate factors dictionary
172 self.factors = {
173 "similarityThreshold": similarityThreshold,
174 "moveRange": moveRange,
175 "letterRange": letterRange,
176 }
178 # Set up agent creator for senders
179 ac_senders = mg.AgentCreator(
180 SenderAgent,
181 model=self,
182 agent_kwargs=self.factors,
183 )
185 # Create agents based on random coordinates generated
186 # in the createData step above, see util.py file.
187 senders = ac_senders.from_GeoDataFrame(
188 initSenderGeoDf,
189 unique_id="unique_id",
190 )
192 # Create random set of initial topic vectors.
193 topics = [
194 tuple(
195 [random.random() for x in range(3)],
196 ) for x in range(self.population)
197 ]
199 # Setup senders
200 for idx, sender in enumerate(senders):
201 # Add to social network
202 self.socialNetwork.add_node(
203 sender.unique_id,
204 numLettersSend=0,
205 numLettersReceived=0,
206 )
207 # Give sender topic
208 sender.topicVec = topics[idx]
209 # Add current topic to dict
210 self.updatedTopicsDict.update(
211 {sender.unique_id: topics[idx]},
212 )
213 # Set random activation weight
214 if useActivation is True:
215 sender.activationWeight = random.random()
216 # Add sender to its region
217 regionID = [
218 x.unique_id for x in self.regions if contains(x.geometry, sender.geometry)
219 ]
220 try:
221 self.space.add_sender(sender, regionID[0])
222 except IndexError as exc:
223 text = f"Problem finding region for {sender.geometry}."
224 raise IndexError(text) from exc
225 # Add sender to schedule
226 self.schedule.add(sender)
228 # Create social network
229 if useSocialNetwork is True:
230 for agent in self.schedule.agents:
231 if isinstance(agent, SenderAgent):
232 self._createSocialEdges(agent, self.socialNetwork)
234 self.datacollector = mesa.DataCollector(
235 model_reporters={
236 "Ledger": getPrunedLedger,
237 "Letters": getScaledLetters ,
238 "Movements": getScaledMovements,
239 "Clusters": getComponents,
240 },
241 )
243 def _createSocialEdges(self, agent: SenderAgent, graph: nx.MultiDiGraph) -> None:
244 """Create social edges with the different wiring factors.
246 Define a close range by using the moveRange parameter. Among
247 these neighbors, create a connection with probability set by
248 the shortRangeNetworkFactor.
250 For all other agents, that are not in this closeRange group,
251 create a connection with the probability set by the longRangeNetworkFactor.
252 """
253 closerange = [x for x in self.space.get_neighbors_within_distance(
254 agent,
255 distance=self.moveRange * self.meandistance,
256 center=False,
257 ) if isinstance(x, SenderAgent)]
258 for neighbor in closerange:
259 if neighbor.unique_id != agent.unique_id:
260 connect = random.choices(
261 population=[True, False],
262 weights=[self.shortRangeNetworkFactor, 1 - self.shortRangeNetworkFactor],
263 k=1,
264 )
265 if connect[0] is True:
266 graph.add_edge(agent.unique_id, neighbor.unique_id, step=0)
267 longrange = [x for x in self.schedule.agents if x not in closerange and isinstance(x, SenderAgent)]
268 for neighbor in longrange:
269 if neighbor.unique_id != agent.unique_id:
270 connect = random.choices(
271 population=[True, False],
272 weights=[self.longRangeNetworkFactor, 1 - self.longRangeNetworkFactor],
273 k=1,
274 )
275 if connect[0] is True:
276 graph.add_edge(agent.unique_id, neighbor.unique_id, step=0)
278 def step(self) -> None:
279 """One simulation step with data collection."""
280 self.step_no_data()
281 self.datacollector.collect(self)
283 def step_no_data(self) -> None:
284 """One simulation step without data collection."""
285 self.scaleSendInput.update(
286 **{x.unique_id: x.numLettersReceived for x in self.schedule.agents},
287 )
288 # Update the currently held topicVec for each agent, based
289 # on potential previouse communication events.
290 for agent in self.schedule.agents:
291 agent.topicVec = self.updatedTopicsDict[agent.unique_id]
292 self.schedule.step()
294 def run(self, n:int) -> None:
295 """Run the model for n steps.
297 Data collection is only run at the end of n steps.
298 This is useful for batch runs accross different
299 parameters.
300 """
301 if self.debug is True:
302 for _ in tqdm(range(n)):
303 self.step_no_data()
304 else:
305 for _ in range(n):
306 self.step_no_data()
307 self.datacollector.collect(self)