Coverage for .tox/p312/lib/python3.10/site-packages/scicom/historicalletters/model.py: 96%

115 statements  

« prev     ^ index     » next       coverage.py v7.4.4, created at 2024-05-28 12:02 +0200

1"""The model class for HistoricalLetters.""" 

2import random 

3from pathlib import Path 

4 

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 

12 

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 

17 

18 

19def getPrunedLedger(model: mesa.Model) -> pd.DataFrame: 

20 """Model reporter for simulation of archiving. 

21 

22 Returns statistics of ledger network of model run 

23 and various iterations of statistics of pruned networks. 

24 

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 

53 

54 

55def getComponents(model: mesa.Model) -> int: 

56 """Model reporter to get number of components. 

57 

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) 

65 

66 

67def getScaledLetters(model: mesa.Model) -> float: 

68 """Return relative number of send letters.""" 

69 return len(model.letterLedger)/model.schedule.time 

70 

71 

72def getScaledMovements(model: mesa.Model) -> float: 

73 """Return relative number of movements.""" 

74 return model.movements/model.schedule.time 

75 

76 

77class HistoricalLetters(mesa.Model): 

78 """A letter sending model with historical informed initital positions. 

79 

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]. 

83 

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. 

88 

89 [1] J. Lobo et al, Population-Area Relationship for Medieval European Cities, 

90 PLoS ONE 11(10): e0162678. 

91 """ 

92 

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__() 

111 

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.updatedPositionDict = {} 

133 self.space = Nuts2Eu() 

134 self.debug = debug 

135 

136 ####### 

137 # Initialize region agents 

138 ####### 

139 

140 # Set up the grid with patches for every NUTS region 

141 # Create region agents 

142 ac = mg.AgentCreator(RegionAgent, model=self) 

143 self.regions = ac.from_file( 

144 regionData, 

145 unique_id="NUTS_ID", 

146 ) 

147 # Add regions to Nuts2Eu geospace 

148 self.space.add_regions(self.regions) 

149 

150 ####### 

151 # Initialize sender agents 

152 ####### 

153 

154 # Draw initial geographic positions of agents 

155 initSenderGeoDf = createData( 

156 population, 

157 populationDistribution=populationDistributionData, 

158 ) 

159 

160 # Calculate mean of mean distances for each agent. 

161 # This is used as a measure for the range of exchanges. 

162 meandistances = [] 

163 for idx in initSenderGeoDf.index.to_numpy(): 

164 name = initSenderGeoDf.loc[idx, "unique_id"] 

165 geom = initSenderGeoDf.loc[idx, "geometry"] 

166 otherAgents = initSenderGeoDf.query(f"unique_id != '{name}'").copy() 

167 geometries = otherAgents.geometry.to_numpy() 

168 distances = [geom.distance(othergeom) for othergeom in geometries] 

169 meandistances.append(mean(distances)) 

170 self.meandistance = mean(meandistances) 

171 

172 # Populate factors dictionary 

173 self.factors = { 

174 "similarityThreshold": similarityThreshold, 

175 "moveRange": moveRange, 

176 "letterRange": letterRange, 

177 } 

178 

179 # Set up agent creator for senders 

180 ac_senders = mg.AgentCreator( 

181 SenderAgent, 

182 model=self, 

183 agent_kwargs=self.factors, 

184 ) 

185 

186 # Create agents based on random coordinates generated 

187 # in the createData step above, see util.py file. 

188 senders = ac_senders.from_GeoDataFrame( 

189 initSenderGeoDf, 

190 unique_id="unique_id", 

191 ) 

192 

193 # Create random set of initial topic vectors. 

194 topics = [ 

195 tuple( 

196 [random.random() for x in range(3)], 

197 ) for x in range(self.population) 

198 ] 

199 

200 # Setup senders 

201 for idx, sender in enumerate(senders): 

202 # Add to social network 

203 self.socialNetwork.add_node( 

204 sender.unique_id, 

205 numLettersSend=0, 

206 numLettersReceived=0, 

207 ) 

208 # Give sender topic 

209 sender.topicVec = topics[idx] 

210 # Add current topic to dict 

211 self.updatedTopicsDict.update( 

212 {sender.unique_id: topics[idx]}, 

213 ) 

214 # Set random activation weight 

215 if useActivation is True: 

216 sender.activationWeight = random.random() 

217 # Add sender to its region 

218 regionID = [ 

219 x.unique_id for x in self.regions if contains(x.geometry, sender.geometry) 

220 ] 

221 try: 

222 self.space.add_sender(sender, regionID[0]) 

223 except IndexError as exc: 

224 text = f"Problem finding region for {sender.geometry}." 

225 raise IndexError(text) from exc 

226 # Prepopulate positions dict 

227 self.updatedPositionDict.update( 

228 {sender.unique_id: [sender.geometry, regionID[0]]}, 

229 ) 

230 # Add sender to schedule 

231 self.schedule.add(sender) 

232 

233 # Create social network 

234 if useSocialNetwork is True: 

235 for agent in self.schedule.agents: 

236 if isinstance(agent, SenderAgent): 

237 self._createSocialEdges(agent, self.socialNetwork) 

238 

239 self.datacollector = mesa.DataCollector( 

240 model_reporters={ 

241 "Ledger": getPrunedLedger, 

242 "Letters": getScaledLetters , 

243 "Movements": getScaledMovements, 

244 "Clusters": getComponents, 

245 }, 

246 ) 

247 

248 def _createSocialEdges(self, agent: SenderAgent, graph: nx.MultiDiGraph) -> None: 

249 """Create social edges with the different wiring factors. 

250 

251 Define a close range by using the moveRange parameter. Among 

252 these neighbors, create a connection with probability set by 

253 the shortRangeNetworkFactor. 

254 

255 For all other agents, that are not in this closeRange group, 

256 create a connection with the probability set by the longRangeNetworkFactor. 

257 """ 

258 closerange = [x for x in self.space.get_neighbors_within_distance( 

259 agent, 

260 distance=self.moveRange * self.meandistance, 

261 center=False, 

262 ) if isinstance(x, SenderAgent)] 

263 for neighbor in closerange: 

264 if neighbor.unique_id != agent.unique_id: 

265 connect = random.choices( 

266 population=[True, False], 

267 weights=[self.shortRangeNetworkFactor, 1 - self.shortRangeNetworkFactor], 

268 k=1, 

269 ) 

270 if connect[0] is True: 

271 graph.add_edge(agent.unique_id, neighbor.unique_id, step=0) 

272 longrange = [x for x in self.schedule.agents if x not in closerange and isinstance(x, SenderAgent)] 

273 for neighbor in longrange: 

274 if neighbor.unique_id != agent.unique_id: 

275 connect = random.choices( 

276 population=[True, False], 

277 weights=[self.longRangeNetworkFactor, 1 - self.longRangeNetworkFactor], 

278 k=1, 

279 ) 

280 if connect[0] is True: 

281 graph.add_edge(agent.unique_id, neighbor.unique_id, step=0) 

282 

283 def step(self) -> None: 

284 """One simulation step with data collection.""" 

285 self.step_no_data() 

286 self.datacollector.collect(self) 

287 

288 def step_no_data(self) -> None: 

289 """One simulation step without data collection.""" 

290 self.scaleSendInput.update( 

291 **{x.unique_id: x.numLettersReceived for x in self.schedule.agents}, 

292 ) 

293 # Update the currently held topicVec for each agent, based 

294 # on potential previouse communication events and the position 

295 # based on a previous movement. 

296 for agent in self.schedule.agents: 

297 agent.topicVec = self.updatedTopicsDict[agent.unique_id] 

298 newGeom = self.updatedPositionDict[agent.unique_id] 

299 if newGeom[0] != agent.geometry: 

300 self.space.move_sender(agent, newGeom[0], newGeom[1]) 

301 self.schedule.step() 

302 

303 def run(self, n:int) -> None: 

304 """Run the model for n steps. 

305 

306 Data collection is only run at the end of n steps. 

307 This is useful for batch runs accross different 

308 parameters. 

309 """ 

310 if self.debug is True: 

311 for _ in tqdm(range(n)): 

312 self.step_no_data() 

313 else: 

314 for _ in range(n): 

315 self.step_no_data() 

316 self.datacollector.collect(self)