import agentpy as ap import pandas as pd import networkx as nx from firm import FirmAgent from product import ProductAgent from orm import db_session, Result import platform import json class Model(ap.Model): def setup(self): self.sample = self.p.sample self.int_stop_times, self.int_stop_t = 0, None self.int_n_iter = int(self.p.n_iter) self.dct_lst_remove_firm_prod = self.p.dct_lst_init_remove_firm_prod self.int_n_max_trial = int(self.p.n_max_trial) self.is_prf_size = bool(self.p.prf_size) self.str_cap_limit_prob_type = str(self.p.cap_limit_prob_type) self.flt_cap_limit_level = float(self.p.cap_limit_level) self.flt_diff_remove = float(self.p.diff_remove) self.proactive_ratio = float(self.p.proactive_ratio) self.int_netw_prf_n = int(self.p.netw_prf_n) # init graph bom G_bom = nx.adjacency_graph(json.loads(self.p.g_bom)) self.product_network = ap.Network(self, G_bom) # init graph firm Firm = pd.read_csv("Firm_amended.csv") Firm['Code'] = Firm['Code'].astype('string') Firm.fillna(0, inplace=True) Firm_attr = Firm.loc[:, ["Code", "Name", "Type_Region", "Revenue_Log"]] firm_product = [] for _, row in Firm.loc[:, '1':].iterrows(): firm_product.append(row[row == 1].index.to_list()) Firm_attr.loc[:, 'Product_Code'] = firm_product Firm_attr.set_index('Code', inplace=True) G_Firm = nx.MultiDiGraph() G_Firm.add_nodes_from(Firm["Code"]) firm_labels_dict = {} for code in G_Firm.nodes: firm_labels_dict[code] = Firm_attr.loc[code].to_dict() nx.set_node_attributes(G_Firm, firm_labels_dict) # init graph firm prod Firm_Prod = pd.read_csv("Firm_amended.csv") Firm_Prod.fillna(0, inplace=True) firm_prod = pd.DataFrame({'bool': Firm_Prod.loc[:, '1':].stack()}) firm_prod = firm_prod[firm_prod['bool'] == 1].reset_index() firm_prod.drop('bool', axis=1, inplace=True) firm_prod.rename({'level_0': 'Firm_Code', 'level_1': 'Product_Code'}, axis=1, inplace=True) firm_prod['Firm_Code'] = firm_prod['Firm_Code'].astype('string') G_FirmProd = nx.MultiDiGraph() G_FirmProd.add_nodes_from(firm_prod.index) firm_prod_labels_dict = {} for code in firm_prod.index: firm_prod_labels_dict[code] = firm_prod.loc[code].to_dict() nx.set_node_attributes(G_FirmProd, firm_prod_labels_dict) # add edge to G_firm according to G_bom for node in nx.nodes(G_Firm): lst_pred_product_code = [] for product_code in G_Firm.nodes[node]['Product_Code']: lst_pred_product_code += list(G_bom.predecessors(product_code)) lst_pred_product_code = list(set(lst_pred_product_code)) # to generate consistant graph lst_pred_product_code = list(sorted(lst_pred_product_code)) for pred_product_code in lst_pred_product_code: # for each product predecessor (component) the firm need # get a list of firm producing this component lst_pred_firm = \ Firm['Code'][Firm[pred_product_code] == 1].to_list() # select multiple supplier (multi-sourcing) n_pred_firm = self.int_netw_prf_n if n_pred_firm > len(lst_pred_firm): n_pred_firm = len(lst_pred_firm) # based on size or not if self.is_prf_size: lst_pred_firm_size = \ [G_Firm.nodes[pred_firm]['Revenue_Log'] for pred_firm in lst_pred_firm] lst_prob = \ [size / sum(lst_pred_firm_size) for size in lst_pred_firm_size] lst_choose_firm = self.nprandom.choice(lst_pred_firm, n_pred_firm, replace=False, p=lst_prob) else: lst_choose_firm = self.nprandom.choice(lst_pred_firm, n_pred_firm, replace=False) lst_add_edge = [(pred_firm, node, {'Product': pred_product_code}) for pred_firm in lst_choose_firm] G_Firm.add_edges_from(lst_add_edge) # graph firm prod set_node_prod_code = set(G_Firm.nodes[node]['Product_Code']) set_pred_succ_code = set(G_bom.successors(pred_product_code)) lst_use_pred_prod_code = list( set_node_prod_code & set_pred_succ_code) for pred_firm in lst_choose_firm: pred_node = [n for n, v in G_FirmProd.nodes(data=True) if v['Firm_Code'] == pred_firm and v['Product_Code'] == pred_product_code][0] for use_pred_prod_code in lst_use_pred_prod_code: current_node = \ [n for n, v in G_FirmProd.nodes(data=True) if v['Firm_Code'] == node and v['Product_Code'] == use_pred_prod_code][0] G_FirmProd.add_edge(pred_node, current_node) # nx.to_pandas_adjacency(G_Firm).to_csv('adj_g_firm.csv') # nx.to_pandas_adjacency(G_FirmProd).to_csv('adj_g_firm_prod.csv') # unconnected node for node in nx.nodes(G_Firm): if G_Firm.degree(node) == 0: for product_code in G_Firm.nodes[node]['Product_Code']: # unconnect node does not have possible suppliers # current node in graph firm prod current_node = \ [n for n, v in G_FirmProd.nodes(data=True) if v['Firm_Code'] == node and v['Product_Code'] == product_code][0] lst_succ_product_code = list( G_bom.successors(product_code)) # different from for different types of product, # finding a common supplier (the logic above), # for different types of product, # finding a custormer for each product for succ_product_code in lst_succ_product_code: # for each product successor (finished product) # the firm sells to, # get a list of firm producing this finished product lst_succ_firm = Firm['Code'][ Firm[succ_product_code] == 1].to_list() # select multiple customer (multi-selling) n_succ_firm = self.int_netw_prf_n if n_succ_firm > len(lst_succ_firm): n_succ_firm = len(lst_succ_firm) # based on size or not if self.is_prf_size: lst_succ_firm_size = \ [G_Firm.nodes[succ_firm]['Revenue_Log'] for succ_firm in lst_succ_firm] lst_prob = \ [size / sum(lst_succ_firm_size) for size in lst_succ_firm_size] lst_choose_firm = \ self.nprandom.choice(lst_succ_firm, n_succ_firm, replace=False, p=lst_prob) else: lst_choose_firm = \ self.nprandom.choice(lst_succ_firm, n_succ_firm, replace=False) lst_add_edge = [(node, succ_firm, {'Product': product_code}) for succ_firm in lst_choose_firm] G_Firm.add_edges_from(lst_add_edge) # graph firm prod for succ_firm in lst_choose_firm: succ_node = \ [n for n, v in G_FirmProd.nodes(data=True) if v['Firm_Code'] == succ_firm and v['Product_Code'] == succ_product_code][0] G_FirmProd.add_edge(current_node, succ_node) self.sample.g_firm = json.dumps(nx.adjacency_data(G_Firm)) self.firm_network = ap.Network(self, G_Firm) self.firm_prod_network = G_FirmProd # import matplotlib.pyplot as plt # nx.draw(G_FirmProd) # plt.show() # init product for ag_node, attr in self.product_network.graph.nodes(data=True): product = ProductAgent(self, code=ag_node.label, name=attr['Name']) self.product_network.add_agents([product], [ag_node]) self.a_lst_total_products = ap.AgentList(self, self.product_network.agents) # init firm for ag_node, attr in self.firm_network.graph.nodes(data=True): firm_agent = FirmAgent( self, code=ag_node.label, name=attr['Name'], type_region=attr['Type_Region'], revenue_log=attr['Revenue_Log'], a_lst_product=self.a_lst_total_products.select([ code in attr['Product_Code'] for code in self.a_lst_total_products.code ])) for product in firm_agent.a_lst_product: # init extra capacity based on discrete uniform distribution assert self.str_cap_limit_prob_type in ['uniform', 'normal'], \ "cap_limit_prob_type other than uniform, normal" if self.str_cap_limit_prob_type == 'uniform': extra_cap_mean = \ firm_agent.revenue_log / self.flt_cap_limit_level extra_cap = self.nprandom.integers(extra_cap_mean-2, extra_cap_mean+2) extra_cap = 0 if round(extra_cap) < 0 else round(extra_cap) # print(firm_agent.name, extra_cap) firm_agent.dct_prod_capacity[product] = extra_cap elif self.str_cap_limit_prob_type == 'normal': extra_cap_mean = \ firm_agent.revenue_log / self.flt_cap_limit_level extra_cap = self.nprandom.normal(extra_cap_mean, 1) extra_cap = 0 if round(extra_cap) < 0 else round(extra_cap) # print(firm_agent.name, extra_cap) firm_agent.dct_prod_capacity[product] = extra_cap self.firm_network.add_agents([firm_agent], [ag_node]) self.a_lst_total_firms = ap.AgentList(self, self.firm_network.agents) # init dct_list_remove_firm_prod (from string to agent) t_dct = {} for firm_code, lst_product in self.dct_lst_remove_firm_prod.items(): firm = self.a_lst_total_firms.select( self.a_lst_total_firms.code == firm_code)[0] t_dct[firm] = self.a_lst_total_products.select([ code in lst_product for code in self.a_lst_total_products.code ]) self.dct_lst_remove_firm_prod = t_dct self.dct_lst_disrupt_firm_prod = t_dct # init output self.lst_dct_lst_remove_firm_prod = [] self.lst_dct_lst_disrupt_firm_prod = [] # set the initial firm product that are removed for firm, a_lst_product in self.dct_lst_remove_firm_prod.items(): for product in a_lst_product: assert product in firm.a_lst_product, \ f"product {product.code} not in firm {firm.code}" firm.a_lst_product_removed.append(product) # proactive strategy # get all the firm prod affected for firm, a_lst_product in self.dct_lst_remove_firm_prod.items(): for product in a_lst_product: init_node = \ [n for n, v in self.firm_prod_network.nodes(data=True) if v['Firm_Code'] == firm.code and v['Product_Code'] == product.code][0] dct_affected = \ nx.dfs_successors(self.firm_prod_network, init_node) lst_affected = set() for i, (u, vs) in enumerate(dct_affected.items()): # at least 2 hops away if i > 0: pred_node = self.firm_prod_network.nodes[u] for v in vs: succ_node = self.firm_prod_network.nodes[v] lst_affected.add((succ_node['Firm_Code'], succ_node['Product_Code'])) lst_affected = list(lst_affected) lst_firm_proactive = \ [lst_affected[i] for i in self.nprandom.choice(range(len(lst_affected)), round(len(lst_affected) * self.proactive_ratio))] for firm_code, prod_code in lst_firm_proactive: pro_firm_prod_code = \ [n for n, v in self.firm_prod_network.nodes(data=True) if v['Firm_Code'] == firm_code and v['Product_Code'] == prod_code][0] pro_firm_prod_node = \ self.firm_prod_network.nodes[pro_firm_prod_code] pro_firm = \ self.a_lst_total_firms.select( [firm.code == pro_firm_prod_node['Firm_Code'] for firm in self.a_lst_total_firms])[0] lst_shortest_path = \ list(nx.all_shortest_paths(self.firm_prod_network, source=init_node, target=pro_firm_prod_code)) dct_drs = {} for di_supp_code in self.firm_prod_network.predecessors( pro_firm_prod_code): di_supp_node = \ self.firm_prod_network.nodes[di_supp_code] di_supp_prod = \ self.a_lst_total_products.select( [product.code == di_supp_node['Product_Code'] for product in self.a_lst_total_products])[0] di_supp_firm = \ self.a_lst_total_firms.select( [firm.code == di_supp_node['Firm_Code'] for firm in self.a_lst_total_firms])[0] lst_cand = self.model.a_lst_total_firms.select([ di_supp_prod in firm.a_lst_product and di_supp_prod not in firm.a_lst_product_removed for firm in self.model.a_lst_total_firms ]) n2n_betweenness = \ sum([True if di_supp_code in path else False for path in lst_shortest_path]) \ / len(lst_shortest_path) drs = n2n_betweenness / \ (len(lst_cand) * di_supp_firm.revenue_log) dct_drs[di_supp_code] = drs dct_drs = dict(sorted( dct_drs.items(), key=lambda kv: kv[1], reverse=True)) for di_supp_code in dct_drs.keys(): di_supp_node = \ self.firm_prod_network.nodes[di_supp_code] di_supp_prod = \ self.a_lst_total_products.select( [product.code == di_supp_node['Product_Code'] for product in self.a_lst_total_products])[0] # find a dfferent firm can produce the same product lst_cand = self.model.a_lst_total_firms.select([ di_supp_prod in firm.a_lst_product and di_supp_prod not in firm.a_lst_product_removed and firm.code != di_supp_node['Firm_Code'] for firm in self.model.a_lst_total_firms ]) if len(lst_cand) > 0: select_cand = self.nprandom.choice(lst_cand) self.firm_network.graph.add_edges_from([ (self.firm_network.positions[select_cand], self.firm_network.positions[pro_firm], { 'Product': di_supp_prod.code }) ]) # print(f"proactive add {select_cand.code} to " # f"{pro_firm.code} " # f"for {di_supp_node['Firm_Code']} " # f"{di_supp_node['Product_Code']}") # change capacity select_cand.dct_prod_capacity[di_supp_prod] -= 1 break # nx.to_pandas_adjacency(G_Firm).to_csv('adj_g_firm_proactive.csv') # draw network # self.draw_network() def update(self): self.a_lst_total_firms.clean_before_time_step() # output self.lst_dct_lst_remove_firm_prod.append( (self.t, self.dct_lst_remove_firm_prod)) self.lst_dct_lst_disrupt_firm_prod.append( (self.t, self.dct_lst_disrupt_firm_prod)) # stop simulation if reached terminal number of iteration if self.t == self.int_n_iter or len( self.dct_lst_remove_firm_prod) == 0: self.int_stop_times = self.t self.stop() def step(self): # print('\n', '=' * 20, 'step', self.t, '=' * 20) # print( # 'dct_list_remove_firm_prod', { # key.name: value.code # for key, value in self.dct_lst_remove_firm_prod.items() # }) # remove_edge_to_cus_and_cus_up_prod for firm, a_lst_product in self.dct_lst_remove_firm_prod.items(): for product in a_lst_product: firm.remove_edge_to_cus_remove_cus_up_prod(product) for n_trial in range(self.int_n_max_trial): # print('=' * 10, 'trial', n_trial, '=' * 10) # seek_alt_supply # shuffle self.a_lst_total_firms self.a_lst_total_firms = self.a_lst_total_firms.shuffle() for firm in self.a_lst_total_firms: if len(firm.a_lst_up_product_removed) > 0: firm.seek_alt_supply() # handle_request # shuffle self.a_lst_total_firms self.a_lst_total_firms = self.a_lst_total_firms.shuffle() for firm in self.a_lst_total_firms: if len(firm.dct_request_prod_from_firm) > 0: firm.handle_request() # reset dct_request_prod_from_firm self.a_lst_total_firms.clean_before_trial() # do not use: # self.a_lst_total_firms.dct_request_prod_from_firm = {} why? # based on a_lst_up_product_removed # update a_lst_product_disrupted / a_lst_product_removed # update dct_lst_disrupt_firm_prod / dct_lst_remove_firm_prod self.dct_lst_remove_firm_prod = {} self.dct_lst_disrupt_firm_prod = {} for firm in self.a_lst_total_firms: if len(firm.a_lst_up_product_removed) > 0: # print(firm.name, 'a_lst_up_product_removed', [ # product.code for product in firm.a_lst_up_product_removed # ]) for product in firm.a_lst_product: n_up_product_removed = 0 for up_product_removed in firm.a_lst_up_product_removed: if product in up_product_removed.a_successors(): n_up_product_removed += 1 if n_up_product_removed == 0: continue else: # update a_lst_product_disrupted # update dct_lst_disrupt_firm_prod if product not in firm.a_lst_product_disrupted: firm.a_lst_product_disrupted.append(product) if firm in self.dct_lst_disrupt_firm_prod.keys(): self.dct_lst_disrupt_firm_prod[firm].append( product) else: self.dct_lst_disrupt_firm_prod[ firm] = ap.AgentList( self.model, [product]) # update a_lst_product_removed # update dct_list_remove_firm_prod # mark disrupted firm as removed based conditionally lost_percent = n_up_product_removed / len( product.a_predecessors()) lst_size = self.a_lst_total_firms.revenue_log lst_size = [firm.revenue_log for firm in self.a_lst_total_firms if product in firm.a_lst_product and product not in firm.a_lst_product_removed ] std_size = (firm.revenue_log - min(lst_size) + 1) / (max(lst_size) - min(lst_size) + 1) prob_remove = 1 - std_size * (1 - lost_percent) # damp prod prob_remove = prob_remove ** self.flt_diff_remove # sample prob prob_remove = self.nprandom.uniform( prob_remove - 0.1, prob_remove + 0.1) prob_remove = 1 if prob_remove > 1 else prob_remove prob_remove = 0 if prob_remove < 0 else prob_remove if self.nprandom.choice([True, False], p=[prob_remove, 1 - prob_remove]): firm.a_lst_product_removed.append(product) if firm in self.dct_lst_remove_firm_prod.keys(): self.dct_lst_remove_firm_prod[firm].append( product) else: self.dct_lst_remove_firm_prod[ firm] = ap.AgentList( self.model, [product]) # print( # 'dct_list_remove_firm_prod', { # key.name: value.code # for key, value in self.dct_lst_remove_firm_prod.items() # }) def end(self): # print('/' * 20, 'output', '/' * 20) # print('dct_list_remove_firm_prod') # for t, dct in self.lst_dct_lst_remove_firm_prod: # for firm, a_lst_product in dct.items(): # for product in a_lst_product: # print(t, firm.name, product.code) # print('dct_lst_disrupt_firm_prod') # for t, dct in self.lst_dct_lst_disrupt_firm_prod: # for firm, a_lst_product in dct.items(): # for product in a_lst_product: # print(t, firm.name, product.code) qry_result = db_session.query(Result).filter_by(s_id=self.sample.id) if qry_result.count() == 0: lst_result_info = [] for t, dct in self.lst_dct_lst_disrupt_firm_prod: for firm, a_lst_product in dct.items(): for product in a_lst_product: db_r = Result(s_id=self.sample.id, id_firm=firm.code, id_product=product.code, ts=t, is_disrupted=True) lst_result_info.append(db_r) db_session.bulk_save_objects(lst_result_info) db_session.commit() for t, dct in self.lst_dct_lst_remove_firm_prod: for firm, a_lst_product in dct.items(): for product in a_lst_product: # only firm disrupted can be removed theoretically qry_f_p = db_session.query(Result).filter( Result.s_id == self.sample.id, Result.id_firm == firm.code, Result.id_product == product.code) if qry_f_p.count() == 1: qry_f_p.update({"is_removed": True}) db_session.commit() self.sample.is_done_flag = 1 self.sample.computer_name = platform.node() self.sample.stop_t = self.int_stop_times db_session.commit() def draw_network(self): import matplotlib.pyplot as plt plt.rcParams['font.sans-serif'] = 'SimHei' pos = nx.nx_agraph.graphviz_layout(self.firm_network.graph, prog="twopi", args="") node_label = nx.get_node_attributes(self.firm_network.graph, 'Name') node_degree = dict(self.firm_network.graph.out_degree()) node_label = { key: f"{key} {node_label[key]} {node_degree[key]}" for key in node_label.keys() } node_size = list( nx.get_node_attributes(self.firm_network.graph, 'Revenue_Log').values()) node_size = list(map(lambda x: x**2, node_size)) edge_label = nx.get_edge_attributes(self.firm_network.graph, "Product") # multi(di)graphs, the keys are 3-tuples edge_label = {(n1, n2): label for (n1, n2, _), label in edge_label.items()} plt.figure(figsize=(12, 12), dpi=300) nx.draw(self.firm_network.graph, pos, node_size=node_size, labels=node_label, font_size=6) nx.draw_networkx_edge_labels(self.firm_network.graph, pos, edge_label, font_size=4) plt.savefig("network.png")