diff --git a/e/recommend_portfolio.py b/e/recommend_portfolio.py index aba537974fa876964ad465515716780f5e4089ee..d15973483314e5cf13133eb419faad69fd2f8e2f 100644 --- a/e/recommend_portfolio.py +++ b/e/recommend_portfolio.py @@ -2,18 +2,16 @@ import sqlite3 import pandas as pd import numpy as np from scipy import optimize -import math import random ## 设定总股票数 -total_stock = 99 +total_stock = 499 ## 风格因子数目 factor_num = 7 ## 寻找组合迭代最大次数 -MAX_ITER_NUM = 100 +MAX_ITER_NUM = 20 - -# %% 从STOCKINFO到STOCK的表名称转换 +#%% 从STOCKINFO到STOCK的表名称转换 def trans_name(info_code): ''' 从STOCKINFO到STOCK的表名称转换 @@ -23,26 +21,26 @@ def trans_name(info_code): SZ000001 ''' letter_num = info_code.split('.') - return letter_num[1] + letter_num[0] + return letter_num[1]+letter_num[0] -# %% （内部函数）获得总股票收益率池 -def get_code_gain(days=200, count=99): +## （内部函数）获得总股票收益率池 +def get_code_gain(days = 200, count = 3670): ''' 计算股票收益率（未排序） :param days: int :param count: int :return: df - 第一列 stockcode 第二列 returns - 共 count（默认200）支股票，returns 求的是days（默认100）天的收益率 + 第一列 stockcode 第二列 returns + 共 count（默认3679）支股票，returns 求的是days（默认100）天的收益率 ''' - conn = sqlite3.connect('data/fin_set.db') - c = conn.cursor() # 创建游标 + conn = sqlite3.connect('fin_set.db') + c = conn.cursor() # 创建游标 code_df = pd.DataFrame( - list(c.execute("SELECT TRADECODE from " + 'STOCKINFO')) + list(c.execute("SELECT TRADECODE from " + 'STOCKINFO')) ) - exist_code_ls = np.array(code_df.iloc[:count, :].T)[0] # 提取前count = 100条股票名称 + exist_code_ls = np.array(code_df.iloc[:count,:].T)[0] #提取前count条股票名称 gain_ls = [] stockcode_ls = [] for info_code in exist_code_ls: @@ -51,57 +49,57 @@ def get_code_gain(days=200, count=99): continue table_code = trans_name(info_code) stockcode_ls.append(table_code) - chg_df = pd.DataFrame(list(c.execute("SELECT CLOSE from " + table_code))) - this_gain = chg_df.iloc[days, 0] - chg_df.iloc[0, 0] + chg_df = pd.DataFrame(list(c.execute("SELECT CLOSE from " +table_code)) ) + this_gain = chg_df.iloc[days,0] - chg_df.iloc[0,0] gain_ls.append(float(this_gain)) - + code_gain = pd.DataFrame({'stockcode': stockcode_ls, 'returns': gain_ls}) return code_gain -# %% （内部函数）获得按收益率排序的总股票池 -def get_sorted_code_gain(days=200, count=99): +## （内部函数）获得按收益率排序的总股票池 +def get_sorted_code_gain(days = 200, count = 3670): ''' 获得按收益率排序的总股票池 :param days: int 选取几日收益率，default 200 :param count: int - 总股票数, default 100 + 总股票数, default 3670 :return: df - 第一列 stockcode 第二列 returns + 第一列 stockcode 第二列 returns ''' code_gain = get_code_gain(days, count) sorted_cg = code_gain.sort_values(by='returns', ascending=False) return sorted_cg -# %% 获得符合收益率档位的股票池 -def get_targeted_code_gain(return_level='low'): +## 获得符合收益率档位的股票池 +def get_targeted_code_gain(return_level = 'low'): ''' 获得符合收益率要求的股票池 :param return_level: str 收益率等级 'low', 'high','mid' :return: dataframe - 第一列 stockcode 第二列 returns + 第一列 stockcode 第二列 returns ''' sorted_cg = get_sorted_code_gain(count=total_stock) count = len(sorted_cg) if return_level == 'low': - return sorted_cg.iloc[-int(count / 3):, :] + return sorted_cg.iloc[-int(count/3):, :] elif return_level == 'high': - return sorted_cg.iloc[:int(count / 3), :] + return sorted_cg.iloc[:int(count/3), :] else: - return sorted_cg.iloc[int(count / 3): 2 * int(count / 3), :] + return sorted_cg.iloc[int(count/3): 2*int(count/3), :] -# %% 最优化问题求解器 +## 最优化问题求解器 def weight_optimizor(stock_gain_df, beta_df, beta_threshold): ''' optimizor :param stock_gain_df: df 组合中的 股票代码+收益率 :param beta_df: df - 约束条件系数矩阵n*m，每列是一只股票（共n值股票），每行是一个因子值（共m个因子）eg. + 约束条件系数矩阵n*m，每列是一只股票（共n值股票），每行是一个因子值（共m个因子）eg. pd.DataFrame({'stock1':[1,2,3],'stock2':[2,3,4]}) :param beta_threshold: 一维 ndarray 限额系数， beta阈值向量 @@ -109,54 +107,54 @@ def weight_optimizor(stock_gain_df, beta_df, beta_threshold): :return: df 两列： stockcode个股代码 weight个股权重 ''' - c = stock_gain_df.iloc[:, 1].values # 目标函数系数 - A_ub = beta_df.values # 系数矩阵，m * n - n, m = A_ub.shape[1], A_ub.shape[0] # 总股票数n,总因子数m - A_eq = np.array([np.zeros(n) + 1]) + c = stock_gain_df.iloc[:,1].values # 目标函数系数 + A_ub = beta_df.values # 系数矩阵，m * n + n,m = A_ub.shape[1], A_ub.shape[0] # 总股票数n,总因子数m + + A_eq = np.array( [np.zeros(n)+1] ) b_eq = np.array([1]) - bounds = (0, 1) + bounds = (0,1) solve = optimize.linprog( - c=-c, A_ub=A_ub, b_ub=beta_threshold, - A_eq=A_eq, b_eq=b_eq, bounds=bounds + c=-c, A_ub=A_ub, b_ub=beta_threshold, + A_eq=A_eq, b_eq= b_eq, bounds=bounds ) - return solve - + return solve -# %% 获取约束矩阵（beta系数矩阵） +## 获取约束矩阵（beta系数矩阵） def get_beta_df(stockcode_ls): ''' 获取约束矩阵（beta系数矩阵） :param list - :return df, 每列是一只股票（共 n 值股票），每行是一个因子值（共 m个因子）eg. + :return df, 每列是一只股票（共 n 值股票），每行是一个因子值（共 m个因子）eg. pd.DataFrame({'stock1':[1,2,3],'stock2':[2,3,4]}) ''' - conn = sqlite3.connect('data/fin_set.db') - c = conn.cursor() # 创建游标 - + conn = sqlite3.connect('fin_set.db') + c = conn.cursor() # 创建游标 + code_beta = pd.DataFrame( list(c.execute("SELECT STOCKCODE,B1,B2,B3,B4,B5,B6,B7 from " + 'BETA')), - columns=['STOCKCODE', 'B1', 'B2', 'B3', 'B4', 'B5', 'B6', 'B7'] - ) - - name_ls = code_beta.iloc[:, 0] + columns=['STOCKCODE','B1','B2','B3','B4','B5','B6','B7'] + ) + + name_ls =code_beta.iloc[:,0] code_ls = [] for info_code in name_ls: table_code = trans_name(info_code) code_ls.append(table_code) code_beta['STOCKCODE'] = code_ls - + filtered_df = code_beta[code_beta['STOCKCODE'].isin(stockcode_ls)].T - - filtered_df.columns = filtered_df.iloc[0, :] - + + filtered_df.columns = filtered_df.iloc[0,:] + filtered_df = filtered_df.iloc[1:, :] return filtered_df -# %% 获取限额向量（beta阈值ndarray）调用此API的时候要给出！！！！ -def get_beta_threshold(beta_threshold=np.full(factor_num, 0.4)): +## 获取限额向量（beta阈值ndarray）调用此API的时候要给出！！！！ +def get_beta_threshold(beta_threshold = np.full(factor_num,0.4)): ''' 获取限额向量（beta阈值ndarray） :param beta_threshold: ndarray @@ -165,45 +163,9 @@ def get_beta_threshold(beta_threshold=np.full(factor_num, 0.4)): np.array([2, 3, 1]) ''' return beta_threshold - - -# %% -target_cg = get_targeted_code_gain(return_level='low').reset_index(drop=True) -stock_num = 10 -total = len(target_cg) -# a = [random.randint(0,total-1) for _ in range(stock_num)] -a = random.sample(range(0, total - 1), stock_num) -print(a) - -stock_gain_df = target_cg.iloc[a, :] -stockcode_ls = np.array(stock_gain_df['stockcode']).tolist() - -conn = sqlite3.connect('data/fin_set.db') -c = conn.cursor() # 创建游标 - -code_beta = pd.DataFrame( - list(c.execute("SELECT STOCKCODE,B1,B2,B3,B4,B5,B6,B7 from " + 'BETA')), - columns=['STOCKCODE', 'B1', 'B2', 'B3', 'B4', 'B5', 'B6', 'B7'] -) - -name_ls = code_beta.iloc[:, 0] -code_ls = [] -for info_code in name_ls: - table_code = trans_name(info_code) - code_ls.append(table_code) -code_beta['STOCKCODE'] = code_ls - -filtered_df = code_beta[code_beta['STOCKCODE'].isin(stockcode_ls)].T - -filtered_df.columns = filtered_df.iloc[0, :] - -filtered_df = filtered_df.iloc[1:, :] - - - + ## 组合中股票数k（默认是10），随机从该收益率档位（默认是收益水平low)的股票池中挑选k个股票 -## 对这k个股票权重进行最优化求解，获得推荐 -def recommend_portfolio(beta_threshold=np.full(factor_num, 0.4), stock_num=10, return_level='low'): +def recommend_portfolio(beta_threshold= np.full(factor_num,0.4), stock_num = 30, return_level = 'low'): ''' 组合中股票数k（默认是10），随机从该收益率档位（默认是收益水平low)的股票池中挑选k个股票 对这k个股票权重进行最优化求解，获得推荐 @@ -219,66 +181,76 @@ def recommend_portfolio(beta_threshold=np.full(factor_num, 0.4), stock_num=10, r 两列 stockcode weight （总 weight是1） ''' - if stock_num > 32: + if stock_num > 165: print('Too many stocks in the portfolio. ') return [] - + target_cg = get_targeted_code_gain(return_level=return_level).reset_index(drop=True) total = len(target_cg) print("total stock num:", total) random.seed(17) success = False iter_num = 0 - while success is False and iter_num < MAX_ITER_NUM: - + possible_portfolios = [] + + # 迭代器 + while iter_num < MAX_ITER_NUM: + # 随机在该收益率档位（默认是收益水平low)的股票池取stock_num支（默认10支）股票+收益率 # 存为stock_gain_df - a = random.sample(range(0, total - 1), stock_num) - stock_gain_df = target_cg.iloc[a, :] - + a = random.sample(range(0, total-1), stock_num) + stock_gain_df = target_cg.iloc[a,:] + # 获取约束矩阵（beta系数） beta_df = get_beta_df(np.array(stock_gain_df['stockcode']).tolist()) # 获取限额向量（阈值beta) beta_threshold = get_beta_threshold(beta_threshold) opt = weight_optimizor(stock_gain_df, beta_df, beta_threshold) - + list = [] if opt.success is True: - success = True - # 转成list输出 + success = True # flag有可行解 + # 转成list for i in range(len(beta_df.columns)): stock = beta_df.columns[i] weight = opt.x[i] if weight != 0: dic = {stock: weight} list.append(dic) - return list - # return pd.DataFrame({'stockcode':beta_df.columns, 'weight': opt.x}) - else: + possible_portfolios.append(list) # 添加进可行解 iter_num = iter_num + 1 - continue - + + # 函数返回 + if success is True: + print("count possible_portfolios:", len(possible_portfolios)) + max_len = 0 + max_id = 0 + for i in range(len(possible_portfolios)): + print("this_len:", len(possible_portfolios[i])) + if len(possible_portfolios[i]) > max_len: + print("max_len: ", max_len) + max_len = len(possible_portfolios[i]) + max_id = i + print("max_len: ", max_len) + return possible_portfolios[max_id] + if success is False or iter_num == MAX_ITER_NUM: - print('Failed to find a portfolio in our pool within ' + str( - MAX_ITER_NUM) + 'iterations,Please adjust your requirements.') + print('Failed to find a portfolio in our pool within '+str(MAX_ITER_NUM)+'iterations,Please adjust your requirements.') return [] # return False - -# %%================ 此api调用示例 ======================= + + + +#%%================ 此api调用示例 ======================= # update: # 输出形式：[{'SZ000006': 0.5025126639003564}, {'SZ000016': 0.14032066376204633}, ... -# 未找到解返回空list # 输出删除了权重为0的股票 - +# rec_portfolio = recommend_portfolio(beta_threshold=np.full(factor_num,0.4), stock_num = 30, return_level = 'mid') # 说明： 均为可选参数，找不到合适解或个股数目要求太多时会提示错误类型 # 提醒：当return_level设为'high'时，阈值最好不低于0.45，否则可能找不到合适解 # beta_threshold 默认np.full(factor_num,0.4) # stock_num 默认30，不超过32 # return_level 默认'low' - -# rec_portfolio = recommend_portfolio(beta_threshold=np.full(factor_num,0.4), stock_num = 10, return_level = 'mid') # print(rec_portfolio) - -