感谢陈轩的《基于机器学习的多因子选股策略》的策略，这篇研究的思路和一些基础函数来源于该策略。
在这里，提取机器学习的基本框架，用基本面因子作为输入，30天后涨幅作为输出，使用随机森林算法。
得到的结果效果并不好，这里仅作为框架分享。

def get_q_Factor(feasible_stocks):q = query(valuation.code, 
          valuation.market_cap,#市值  valuation.circulating_market_cap,  balance.total_assets - balance.total_liability,#净资产  balance.total_assets / balance.total_liability, 
          indicator.net_profit_to_total_revenue, #净利润/营业总收入  indicator.inc_revenue_year_on_year,  #营业收入增长率（同比）  balance.development_expenditure, #RD  valuation.pe_ratio, #市盈率（TTM）  valuation.pb_ratio, #市净率（TTM）  indicator.inc_net_profit_year_on_year,#净利润增长率（同比）  balance.dividend_payable,  indicator.roe,  indicator.roa,  income.operating_profit / income.total_profit, #OPTP  indicator.gross_profit_margin, #销售毛利率GPM  balance.fixed_assets / balance.total_assets, #FACR  valuation.pcf_ratio, #CFP  valuation.ps_ratio #PS).filter(valuation.code.in_(feasible_stocks))return qdef initialize_df(df):#定义列名df.columns = ['code', 'mcap', 'CMV','log_NC', 'LEV', 'NI_p', 'g', 'development_expenditure','pe','BP','G_p','dividend_payable','ROE','ROA','OPTP','GPM','FACR','pcf_ratio','PS']#标签：对数市值df['log_mcap'] = np.log(df['mcap'])#因子：df['EP'] = df['pe'].apply(lambda x: 1/x)df['BP'] = df['BP'].apply(lambda x: 1/x)df['DP'] = df['dividend_payable']/(df['mcap']*100000000)#因子：df['RD'] = df['development_expenditure']/(df['mcap']*100000000)# 因子：现金收益率df['CFP'] = df['pcf_ratio'].apply(lambda x: 1/x)df['log_NC'] = np.log(df['log_NC'])df['CMV'] = np.log(df['CMV'])#因子：净利润率df['NI_p'] = np.abs(df['NI_p'])#因子：df['NI_n'] = np.abs(df['NI_p'][df['NI_p']<0])   df['PEG'] = df['pe'] / (df['G_p']*100)del df['mcap']del df['pe']del df['dividend_payable']del df['pcf_ratio']del df['development_expenditure']df = df.fillna(0)return df__standardizeList = ['log_mcap','log_NC', 'LEV', 'NI_p', 'NI_n', 'g', 'RD','EP','BP','G_p','PEG','DP','CMV','ROE','ROA','OPTP','GPM','FACR','CFP','PS']

secCode = '000985.XSHG'sample = get_index_stocks(secCode)#print(len(sample))#print(sample)feasible_stocks = sampleq_Factor = get_q_Factor(feasible_stocks)#print(q_Factor)write_file("test.txt", str(q_Factor))from datetime import datetime, timedelta ,datefrom jqfactor import winsorize_medimport jsondate1 = datetime(2018,3,26)df_train = get_fundamentals(q_Factor, date = date1)df_train = initialize_df(df_train)write_file("机器学习/test.txt", str(df_train))write_file('机器学习/df_train.csv', df_train.to_csv(), append=False)from jqfactor import standardlizefor fac in __standardizeList:df_train[fac] = winsorize_med(df_train[fac], scale=5, inclusive=True, inf2nan=True, axis=0)    df_train[fac] = standardlize(df_train[fac], inf2nan=True, axis=0)write_file('df_train_std.csv', df_train.to_csv(), append=False)

/opt/conda/lib/python3.6/site-packages/jqdata/db_utils.py:234: SADeprecationWarning: Compiled objects now compile within the constructor.
  comp.compile()
/opt/conda/lib/python3.6/site-packages/ipykernel_launcher.py:63: RuntimeWarning: invalid value encountered in log

1170397

import pandas as pd#获取多支股票的收盘价pan = get_price(sample, start_date=date1, end_date=date1+timedelta(30), \               frequency='1d', fields=['close'])close = pan['close']write_file('机器学习/close.csv', close.to_csv(), append=False)#从收盘价获取30天涨幅closefirst = close.iloc[0]closelast = close.iloc[-1]rate = (closelast - closefirst)/closefirstrate = pd.DataFrame(rate,columns=['rate'])#具有财务指标的股票数量少于具有价格的股票数量，所以需要滤除rate = rate[rate.index.isin(list(df_train['code']))]write_file('机器学习/rate.csv', rate.to_csv(), append=False)write_file('机器学习/close.csv', close.to_csv(), append=False)returns = rate['rate']returns = winsorize_med(returns, scale=5, inclusive=True, inf2nan=True, axis=0)  returns = standardlize(returns, inf2nan=True, axis=0)returns = returns.fillna(0)from sklearn.ensemble import RandomForestRegressormodel = RandomForestRegressor(random_state=42,n_estimators=500,n_jobs=-1)__trainList = ['log_mcap','log_NC', 'LEV', 'NI_p','g', 'EP','BP','G_p','PEG','CMV','ROE','ROA','OPTP','GPM','FACR','CFP', 'PS']__trainList = ['BP']#trainval = df_train[__trainList]write_file('机器学习/trainval.csv', trainval.to_csv(), append=False)model.fit(trainval,returns)

/opt/conda/lib/python3.6/site-packages/jqresearch/api.py:87: FutureWarning: 
Panel is deprecated and will be removed in a future version.
The recommended way to represent these types of 3-dimensional data are with a MultiIndex on a DataFrame, via the Panel.to_frame() method
Alternatively, you can use the xarray package http://xarray.pydata.org/en/stable/.
Pandas provides a `.to_xarray()` method to help automate this conversion.

  pre_factor_ref_date=_get_today())

RandomForestRegressor(bootstrap=True, criterion='mse', max_depth=None,
           max_features='auto', max_leaf_nodes=None,
           min_impurity_decrease=0.0, min_impurity_split=None,
           min_samples_leaf=1, min_samples_split=2,
           min_weight_fraction_leaf=0.0, n_estimators=500, n_jobs=-1,
           oob_score=False, random_state=42, verbose=0, warm_start=False)

secCode = '000985.XSHG'sample = get_index_stocks(secCode)feasible_stocks = sampleq_Factor = get_q_Factor(feasible_stocks)from datetime import datetime, timedelta ,datefrom jqfactor import winsorize_medimport jsondate2 = datetime(2018,5,24)df_test = get_fundamentals(q_Factor, date = date2)df_test = initialize_df(df_test)print(df_test.index)from jqfactor import standardlizefor fac in __standardizeList:df_train[fac] = winsorize_med(df_test[fac], scale=5, inclusive=True, inf2nan=True, axis=0)    df_train[fac] = standardlize(df_test[fac], inf2nan=True, axis=0)write_file('df_test_std.csv', df_test.to_csv(), append=False)testval = df_test[__trainList]returns_test = pd.DataFrame(model.predict(testval),columns=['returns_test'],index=df_test['code'])write_file('机器学习/return_test.csv', returns_test.to_csv(), append=False)returns_test.sort_values(by = ['returns_test'],axis = 0,ascending = False)

/opt/conda/lib/python3.6/site-packages/jqdata/db_utils.py:234: SADeprecationWarning: Compiled objects now compile within the constructor.
  comp.compile()
/opt/conda/lib/python3.6/site-packages/ipykernel_launcher.py:63: RuntimeWarning: invalid value encountered in log

RangeIndex(start=0, stop=3407, step=1)

.dataframe tbody tr th:only-of-type {        vertical-align: middle;    }    .dataframe tbody tr th {        vertical-align: top;    }    .dataframe thead th {        text-align: right;    }

3407 rows × 1 columns