mesa/computation.py

import json
import os
import datetime

import networkx as nx
import pandas as pd
from mesa import Model

from typing import TYPE_CHECKING

from my_model import MyModel

if TYPE_CHECKING:
    from controller_db import ControllerDB


class Computation:

    def __init__(self, c_db: 'ControllerDB'):
        # 控制不同进程 计算不同的样本 但使用同一个 数据库 c_db
        self.c_db = c_db
        self.pid = os.getpid()

    def run(self, str_code='0', s_id=None):
        sample_random = self.c_db.fetch_a_sample(s_id)
        if sample_random is None:
            return True

        # lock this row by update is_done_flag to 0 将运行后的样本设置为 flag 0
        self.c_db.lock_the_sample(sample_random)
        print(
            f"Pid {self.pid} ({str_code}) is running "
            f"sample {sample_random.id} at {datetime.datetime.now()}")
        # 将sample 对应的 experiment 的一系列值 和 参数值 传入 模型 中 包括列名 和 值
        dct_exp = {column: getattr(sample_random.experiment, column)
                   for column in sample_random.experiment.__table__.c.keys()}
        # 删除不需要的 主键
        del dct_exp['id']

        dct_sample_para = {'sample': sample_random,
                           'seed': sample_random.seed,
                           **dct_exp}

        product_network_test = nx.adjacency_graph(json.loads(dct_sample_para['g_bom']))

        model = MyModel(dct_sample_para)
        for i in range(1):
            model.step()
            print(i, datetime.datetime.now())
        model.end()
        return False

    def initialize_firm_network(self):
        # Read the firm data

        firm = pd.read_csv("input_data/input_firm_data/Firm_amended.csv")

        firm['Code'] = firm['Code'].astype(str)

        firm.fillna(0, inplace=True)

        firm_attr = firm.loc[:, ["Code", "Type_Region", "Revenue_Log"]]

        firm_industry_relation = pd.read_csv("input_data/firm_industry_relation.csv")
        firm_industry_relation['Firm_Code'] = firm_industry_relation['Firm_Code'].astype('string')

        firm_product = []

        grouped = firm_industry_relation.groupby('Firm_Code')['Product_Code'].apply(list)
        firm_product.append(grouped)

        firm_attr['Product_Code'] = firm_attr['Code'].map(grouped)
        firm_attr.set_index('Code', inplace=True)

        grouped = firm_industry_relation.groupby('Firm_Code')
        self.firm_prod_labels_dict = {code: group['Product_Code'].tolist() for code, group in grouped}

        # 遍历'Product_Code' 与 index 交换
        for index, row in firm_attr.iterrows():
            id_index_list = []
            for i in row['Product_Code']:
                for key_values in self.id_code.items():
                    if int(key_values[0]) == i:
                        for id in key_values[1]:
                            id_index_list.append(id)
                        firm_attr.at[index, 'Product_Code'] = id_index_list

        self.G_Firm.add_nodes_from(firm["Code"])
        # Assign attributes to the firm nodes
        firm_labels_dict = {code: firm_attr.loc[code].to_dict() for code in self.G_Firm.nodes}
        nx.set_node_attributes(self.G_Firm, firm_labels_dict)

        self.Firm = firm

    def initialize_firm_product_network(self):

        firm_industry_relation = pd.read_csv("input_data/firm_industry_relation.csv")
        firm_industry_relation['Firm_Code'] = firm_industry_relation['Firm_Code'].astype('string')
        firm_industry_relation['Product_Code'] = firm_industry_relation['Product_Code'].apply(lambda x: [x])
        # 将 'firm_prod' 表中的每一行作为图中的节点
        self.G_FirmProd.add_nodes_from(firm_industry_relation.index)
        # 为每个节点分配属性

        # 遍历'Product_Code' 与 index 交换
        for index, row in firm_industry_relation.iterrows():
            id_index_list = []
            for i in row['Product_Code']:
                for key_values in self.id_code.items():
                    if int(key_values[0]) == i:
                        for id in key_values[1]:
                            id_index_list.append(id)
                        firm_industry_relation.at[index, 'Product_Code'] = id_index_list

        firm_prod_labels_dict = {code: firm_industry_relation.loc[code].to_dict() for code in
                                 firm_industry_relation.index}
        nx.set_node_attributes(self.G_FirmProd, firm_prod_labels_dict)

    def add_edges_to_firm_network(self):
        """ Add edges between firms based on the product BOM relationships """
        # Add edges to G_Firm according to G_bom
        for node in nx.nodes(self.G_Firm):
            lst_pred_product_code = []
            for product_code in self.G_Firm.nodes[node]['Product_Code']:
                lst_pred_product_code += list(self.G_bom.predecessors(product_code))
            lst_pred_product_code = list(set(lst_pred_product_code))
            lst_pred_product_code = list(sorted(lst_pred_product_code))  # Ensure consistency

            for pred_product_code in lst_pred_product_code:
                # Get a list of firms producing the component (pred_product_code)
                lst_pred_firm = [firm_code for firm_code, product in self.firm_prod_labels_dict.items() if
                                 pred_product_code in product]

                # Select multiple suppliers (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)

                if self.is_prf_size:
                    # 获取 firm 的 size 列表
                    lst_pred_firm_size = [self.G_Firm.nodes[pred_firm]['Revenue_Log'] for pred_firm in lst_pred_firm]
                    # 检查 lst_pred_firm_size 是否为空或总和为 0
                    if len(lst_pred_firm_size) == 0 or sum(lst_pred_firm_size) == 0:
                        # print("警告: lst_pred_firm_size 为空或总和为 0，无法生成概率分布")
                        lst_choose_firm = []  # 返回空结果，或根据需要处理
                    else:
                        # 计算总和
                        sum_pred_firm_size = sum(lst_pred_firm_size)
                        # 归一化生成 lst_prob
                        lst_prob = [size / sum_pred_firm_size for size in lst_pred_firm_size]
                        # 使用 np.isclose() 确保概率总和接近 1
                        if not np.isclose(sum(lst_prob), 1.0):
                            # print(f"警告: 概率总和为 {sum(lst_prob)}，现在进行修正")
                            lst_prob = [prob / sum(lst_prob) for prob in lst_prob]
                        # 确保没有负值或 0
                        lst_prob = [max(0, prob) for prob in lst_prob]
                        # 根据修正后的概率选择 firm
                        lst_choose_firm = self.nprandom.choice(lst_pred_firm, n_pred_firm, replace=False, p=lst_prob)
                else:
                    # 直接进行随机选择
                    lst_choose_firm = self.nprandom.choice(lst_pred_firm, n_pred_firm, replace=False)

                # Add edges from predecessor firms to current node (firm)
                lst_add_edge = [(pred_firm, node, {'Product': pred_product_code}) for pred_firm in lst_choose_firm]
                self.G_Firm.add_edges_from(lst_add_edge)

                # Add edges to firm-product network
                self.add_edges_to_firm_product_network(node, pred_product_code, lst_choose_firm)

    def add_edges_to_firm_product_network(self, node, pred_product_code, lst_choose_firm):
        """ Helper function to add edges to the firm-product network """
        set_node_prod_code = set(self.G_Firm.nodes[node]['Product_Code'])
        set_pred_succ_code = set(self.G_bom.successors(pred_product_code))
        lst_use_pred_prod_code = list(set_node_prod_code & set_pred_succ_code)

        if len(lst_use_pred_prod_code) == 0:
            print("错误")

        pred_node_list = []
        for pred_firm in lst_choose_firm:
            for n, v in self.G_FirmProd.nodes(data=True):
                for v1 in v['Product_Code']:
                    if v1 == pred_product_code and v['Firm_Code'] == pred_firm:
                        pred_node_list.append(n)
            if len(pred_node_list) != 0:
                pred_node = pred_node_list[0]
            else:
                pred_node = -1
            current_node_list = []
            for use_pred_prod_code in lst_use_pred_prod_code:
                for n, v in self.G_FirmProd.nodes(data=True):
                    for v1 in v['Product_Code']:
                        if v1 == use_pred_prod_code and v['Firm_Code'] == node:
                            current_node_list.append(n)
            if len(current_node_list) != 0:
                current_node = current_node_list[0]
            else:
                current_node = -1
            if current_node != -1 and pred_node != -1:
                self.G_FirmProd.add_edge(pred_node, current_node)

    def connect_unconnected_nodes(self):
        """ Connect unconnected nodes in the firm network """
        for node in nx.nodes(self.G_Firm):
            if self.G_Firm.degree(node) == 0:
                current_node_list = []
                for product_code in self.G_Firm.nodes[node]['Product_Code']:
                    for n, v in self.G_FirmProd.nodes(data=True):
                        for v1 in v['Product_Code']:
                            if v['Firm_Code'] == node and v1 == product_code:
                                current_node_list.append(n)
                    if len(current_node_list) != 0:
                        current_node = current_node_list[0]
                    else:
                        current_node = -1
                    lst_succ_product_code = list(self.G_bom.successors(product_code))

                    for succ_product_code in lst_succ_product_code:
                        lst_succ_firm = [firm_code for firm_code, product in self.firm_prod_labels_dict.items() if
                                         succ_product_code in product]

                        n_succ_firm = self.int_netw_prf_n
                        if n_succ_firm > len(lst_succ_firm):
                            n_succ_firm = len(lst_succ_firm)

                        if self.is_prf_size:
                            lst_succ_firm_size = [self.G_Firm.nodes[succ_firm]['Revenue_Log'] for succ_firm in
                                                  lst_succ_firm]
                            if len(lst_succ_firm_size) == 0 or sum(lst_succ_firm_size) == 0:
                                # print("警告: lst_pred_firm_size 为空或总和为 0，无法生成概率分布")
                                lst_choose_firm = []  # 返回空结果，或根据需要处理
                            else:
                                # 计算总和
                                sum_pred_firm_size = sum(lst_succ_firm_size)
                                # 归一化生成 lst_prob
                                lst_prob = [size / sum_pred_firm_size for size in lst_succ_firm_size]
                                # 使用 np.isclose() 确保概率总和接近 1
                                if not np.isclose(sum(lst_prob), 1.0):
                                    # print(f"警告: 概率总和为 {sum(lst_prob)}，现在进行修正")
                                    lst_prob = [prob / sum(lst_prob) for prob in lst_prob]

                                # 确保没有负值或 0
                                lst_prob = [max(0, prob) for prob in lst_prob]

                                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]
                        self.G_Firm.add_edges_from(lst_add_edge)

                        # Add edges to firm-product network
                        succ_node_list = []
                        for succ_firm in lst_choose_firm:
                            for n, v in self.G_FirmProd.nodes(data=True):
                                for v1 in v['Product_Code']:
                                    if v1 == succ_product_code and v['Firm_Code'] == succ_firm:
                                        succ_node_list.append(n)
                            if len(succ_node_list) != 0:
                                succ_node = succ_node_list[0]
                            else:
                                succ_node = -1

                            if current_node != -1 and succ_node != -1:
                                self.G_FirmProd.add_edge(current_node, succ_node)

        self.sample.g_firm = json.dumps(nx.adjacency_data(self.G_Firm))
        self.firm_network = self.G_Firm  # 直接使用 networkx 图对象
        self.firm_prod_network = self.G_FirmProd  # 直接使用 networkx 图对象