Agent-based modeling 的電腦實驗，最核心的架構不外乎一個大迴圈（super loop）包裹著一群規則。大迴圈每跑一輪，系統就更新一次狀態，就如同時鐘的滴答（tick）聲般。通常系統每次滴答都會收集一次統計資料。這類實驗，有許多現成的 famework 可用，如最經典的 Swarm 及其後進 Repast ，還有我模仿 Repast ，自己搞的一個 ，它們都提供了 start, pause, stop 等流程控制的介面。
模擬複雜網路，也可以套用 Agent-based modeling 架構。不過諸如網路的群聚度（clustering coefficient, C）及網路特徵的路經長度（characteristic path length, L）等統計數據計算需耗費的時間，隨著網路的規模成長很快，所以不適合運作太頻繁。但我們卻得靠這些統計數據來判斷網路是否收斂、試驗是否該終止了，想試驗各個參數排列組合時，更是雪上加霜。
透過 Intermediate File 是很直覺的解法。程式每次執行都餵入一個輸入檔來初始網路結構，並以參數決定 super loop 要跑幾次。統計數據只在程式要結束時計算，然後將網路狀態及統計數據吐給輸出檔。下次要執行，就以這個輸出檔當作新的輸入，然後一樣決定要跑幾圈，最後得到另一個輸出檔。採取接力的方式，不用每次都從頭跑，不會浪費先前程式執行的時間。
Multithread 是另一個可行作法。採用 Agent-based modeling framework 的方式，一個 thread 跑核心的規則。另一個 thread 則控制試驗的流程，讓 user 決定何時該暫停下來，計算統計數據，然後決定是否繼續執行。
Dynamic Programming Language 是這次的正解。利用動態程式語言，如 Python ，可以很方便於執行時期控制模擬試驗的程式流程。我就是採用 Python ，搭配 NetworkX 來處理 Graph ，並以 matplotlib 來繪製圖表。接下來就看看如何架設這個試驗平台：
- 下載 Python Enthought Edition 的 installer 。它是 Python 的加強版，除了 Python 基本工具外，還整合了許多有用的工具，例如 matplotlib 。
- 到這裡下載 NetworkX 的 installer 。
- 依序執行這兩個 installer 。
裝好了後，就以 Emergence of a small world from local interactions: modeling acquaintance networks 裡描述的實驗來操刀，試寫第一個 Python 程式：
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 |
""" Simulation for the verifying of "Emergence of a Small World from Local Interactions: Modeling Acquaintance Network", Davidsen J. et al. 2002. Public Variables: - Iters: the interations of loop, used in loop() Private Variables: - _G: graph of the acquaintance network - _N: the total of nodes - _p: the death-and-birth probability, used in _step2() - _run: a generator returned by loop() Usage ===== This module must run on a python shell with interative mode; and call public functions e.g. runnext(). Example ------- You can simply run this module with default settings. > python -i acq2002.py >>> runnext(10) # loops 10 times of Iters time-steps. ... Setting parameters is allowed. >>> start(1000, 0.04) # starts the simulation with new N, and p >>> Iters=10000 # sets the iterations of each loop >>> runnext(5) # loops 5 times of 10000 time-steps. ... >>> Iters=100000 # sets the iterations of each loop >>> runnext(2) # loops another 2 times of 100000 time-steps ... After above, you may want to save the result. >>> savepkfig() >>> savenet() """ __author__ = "Jiang Yu-Kuan, yukuan.jiang(at)gmail.com" __date__ = "2006/08/22~2006/10/02" __revision__ = "1.10" import random as rnd import networkx as nx # Must import this as a name to avoid namespace collision! def _create(): """Create an empty acquaintance network. Adds _N nodes to _G, and _G contains no any edges. """ global _G _G = nx.Graph() _G.add_nodes_from(xrange(_N)) # adds n nodes into _G def _step1(): """Friend making of two persons. One randomly chosen person picks any two his acquaintances and introduces them to each another. If they have not met before, a new link between them is formed. In case the person chosen has less than two acquaintances, he introduces himself to one other random person. """ u, v = rnd.sample(xrange(_N), 2) nb = _G.neighbors(u) if len(nb) > 1: u, v = rnd.sample(nb, 2) _G.add_edge(u, v) def _step2(p): """Death and Birth of a chosen person. With probability p, one randomly chosen person is removed from the network, including all links connected to this node, and replaced by a new person with one randomly chosen acquaintance. """ if rnd.random() < p: # sol. 1 v = rnd.randrange(_N) _G.delete_node(v) _G.add_node(v) # sol. 2: slower than sol. 1. #_G.delete_edges_from(_G.edges(rnd.randrange(_N))) def _avgK2(): """Return the average square of degree of all nodes.""" kss = [k*k for k in _G.degree(_G.nodes())] return sum(kss)/float(_N) def _avgSPL(): """Return the average shortest path length of Graph _G.""" pathlengths=[] for v in _G.nodes(): # sol. 1 #spl = nx.single_source_shortest_path_length(_G,v) # including v to v #for pl in spl.values(): # pathlengths.append(pl) # sol. 2 pathlengths += nx.single_source_shortest_path_length(_G,v).values() return sum(pathlengths) / float(len(pathlengths)-_N) def _stat(): """Gather statistics for Graph _G.""" print "N, p = %d, %f" % (_N, _p) print "Degree histogram:", nx.degree_histogram(_G) print "Avg. degree, <k>:", 2.*_G.number_of_edges()/_N print "Avg. square of degree, <k^2>:", _avgK2() print "Avg. clustering coefficient, C:", nx.average_clustering(_G) print "Avg. shortest path length, L:", _avgSPL() # spends huge time def savepkfig(fn='acq2002pk.png'): """Save the p(k) figure. - fn: file name """ import pylab as mpl dh = nx.degree_histogram(_G) pk = mpl.array(dh)/float(_N) mpl.loglog(pk, 'r--') mpl.grid(True) mpl.gca().xaxis.grid(True, which='minor') mpl.xlabel('k') mpl.ylabel('p(k)') mpl.title('N=%d, p=%d' % (_N, _p) ) mpl.savefig(fn) def savenet(fn='acq2002', fm='GPickle'): """Save the acquaintance network. - fn: file name - fm: file format """ save = {'edgelist':nx.write_edgelist, # edge list 'adjlist':nx.write_adjlist, # node adjacency-list 'yaml':nx.write_yaml, # YAML text format 'gpickle':nx.write_gpickle # Python pickle format. } fm = fm.lower() if fm not in save.keys(): fm = 'edgelist' import os fn, ext = os.path.splitext(fn) ext = ext[1:].lower() if ext in save.keys(): fm = ext save[fm](_G, '.'.join([fn, fm])) def _loop(): """Loop _step1 and _step2 of Iters times and gather statistics. This function uses yield statement and returns the total iterations of loop """ import time t_start = time.time() # records the start time t_ttl = 0 # clears the total time spent i_cnt = i_ttl = 0 # clears the loop counter and total loops. _create() # create an empty network while 1: i_cnt += 1 _step1() _step2(_p) if i_cnt == Iters: i_ttl += i_cnt i_cnt = 0 _stat() t_inc = time.time() - t_start t_ttl += t_inc print "Time spent (increment,total): (%f,%f)" % (t_inc, t_ttl) yield i_ttl t_start = time.time() def runnext(n=1): """Call _run.next() of n times.""" for i in xrange(n): print _run.next() def start(N=100, p=0.04): """Start the module. - N: the total of nodes of the network - p: the death-and-birth probability """ global _N, _p, _run _N, _p = N, p _run = _loop() # gets a generator Iters = 100000 if __name__ == "__main__": # Uses Psyco if available try: import psyco psyco.full() print "Psyco: OK" except ImportError: print "Psyco: FAIL" start(100, .04) """ import profile, pstats profile.run('runnext()', 'acq2002.prof') s = pstats.Stats('acq2002.prof') s.sort_stats('time','name').print_stats() """ |
- 行 203~211 的 start() 除了測試這個模組外，也用作主程式。
- 行 211 讓我們取得 generator 給 _run 。
- 行 214 設定每次 _run.next() 要跑幾輪 super loop ，每 _run.next() 一次，才統計一次數據。
- 行 225 執行 start() ，並指定不同的 node 數 N 及死生發生的機率 p 。
- 這個模組執行後，可以在 Python shell 下 _run.next() ，每 _run.next() 一次，大迴圈就跑 Iters 輪。此外，還可以在 Python shell 中改變 Iters 的次數。
- 行 197~200 的 runnext() 讓 Python shell 下，調用多次 _run.next() 更方便。
- 行 217~223 使用 Psyco 來加速。
註：
- NetworkX 的用法可以參考其 Tutorial 及 API Reference 。
- Python 的 random 模組說明可以參考這裡 。
- 要以 matplotlib 繪製圖表，可以參考其 Tutorial 及 Screenshots 。其刻意模仿 Matlab 的用法，用來倍感親切。
- 其他 Python 的說明文件可以參考 Python Tutorials 。
