pd.qcut, pd.cut, df.groupby()等在分组和聚合方面的应用

量化交易里, 需要进行大量的分组和统计, 以方便自己处优势的位置/机会.

比如对股价进行趋势分析, 波动性分析, 量化之后, 进行归类统计, 再进行胜算概率的统计.

依据D8和T8的区间, 能够组合出来16种情形, 每一种case都是人们搭建起来的一幅美丽的场景.

研究和观察每一幅场景出现以后, 随后的几天里的表现, 那是非常令人期待的事情.

TD图的统计展示, 还没有做完, 先留下部分结果:

代码:

def study_t_d_map():

    stk=load_symbol('000895.sz', start='2016-01-01')

    c=stk.ohlc.close

    dft=pd.DataFrame()

    n=32

    dft['c']=c

    #dft['std']  = ttr.stddev(c,32)

    #dft['std_'] = c.rolling(32).std()

    #dft['lr'] = ttr.linreg(c,n)

    #slope = ta.LINEARREG_SLOPE(c,n)

    dft['trendNorm'] = trendNorm(c,32)

    dft['trend16'] = trendNorm(c,16)

    dft['trend8'] = trendNorm(c,8)

    beta = ttr.estimateBeta(dft.trend16+15, dft.trend8+15)

    print('beta(T16, T8): ', beta) # 0.10

    fig,ax = plt.subplots(1,1)

    ax.scatter(dft.trend8, dft.trend16)

    dft['v32'] =   ttr.stddev(c,32)/ ta.LINEARREG(c,32)

    dft['v16'] =   ttr.stddev(c,16)/ ta.LINEARREG(c,16)

    dft['v8'] =   ttr.stddev(c,8)/ ta.LINEARREG(c,8)

    dft['D']  = (c-ttr.linreg(c,n) )/ ta.STDDEV(c,n)

    n=16; dft['D16']  = (c-ttr.linreg(c,n) )/ ta.STDDEV(c,n)

    n=8; dft['D8']  = (c-ttr.linreg(c,n) )/ ta.STDDEV(c,n)

    print('==>数据的长度是: {}'.format(len(c)))

    beta2 = ttr.estimateBeta(dft.D16+5, dft.D8+5)

    print('beta(D16, D8): ', beta2) # 0.67

    fig,(ax1,ax2) = plt.subplots(1,2)

    ax1.scatter(dft.D8, dft.D16)

    ax2.scatter(dft.D8, dft.trend8)

    ax2.set_title('D8 - T8 map')

    beta3 = ttr.estimateBeta(dft.D8+15, dft.trend8+15)

    print('beta(D8, T8): ', beta3)  # 0.23

    roc2 = ttr.roc(c,2).shift(-2)

    D8 = dft['D8']

    T8 = dft['trend8']

    #whichCut=value_cut # quantile_cut

    whichCut= 'cut'

    if whichCut== 'qcut':

        cut=quantile=[0,0.25,0.5,0.75,1]

        d_key = pd.qcut(D8, q=quantile,

                        labels=['BDL', 'NDL','NDR','BDR'])

        t_key = pd.qcut(T8, q=quantile,

                        labels=['BTD', 'NTD','NTU','BTU'])

    else:

        cut=[-8,-1,0 ,1,8]

        d_key = pd.cut(D8, bins=cut)

        t_key = pd.cut(T8, bins=cut)

    r=pd.DataFrame()

    #r['c'] = c

    r['roc2'] = roc2.dropna()

    r['d_key'] = d_key

    r['t_key'] = t_key

    def count_positive_returns(x):

        count,countp,countn=0,0,0

        cumr=1.0

        for e in x:

            cumr *= (1.+e)

            count += 1

            if e>=0:

                countp += 1

            else:

                countn += 1

        return ( count, countp, countn ,round(cumr,3)  )

    groupby = r['roc2'].groupby([r['d_key'], r['t_key']])

    tot_count = groupby.count()

    p_count = groupby.apply(count_positive_returns)

    p_count

    #把一片一片的分组数据做成 字典

    pian = OrderedDict(list(groupby))

    D8T8, count_list= [],[]

    for k in pian.keys():

#    pian[('BDL','BTD')]

        len_=len(pian[k])

        print(len_)

        count_list.append(len_)

        D8T8.append( pd.DataFrame({'T8':T8,'D8':D8}, index=pian[k].index))

    fig,ax=plt.subplots(1,1)

    y=x=[-1.5,-0.5,0, 1.5,2.5]

    for i,_df in enumerate(D8T8):

        ax.scatter(_df.D8, _df.T8, label='Q' + str(i+1)+\

                   ':'+str(count_list[i]))

#    for row in range(4):

#        for col in range(4):

#            print (x[row], y[col])

#            pos=row*col

#            print(count_list[pos])

#        ax.text(x,y, count_list[i])

    ax.set_title('D8 - T8 map by cut:'+str(cut))

    ax.legend()

    xticks = [-2,-1. ,  0. ,   1. ,2]

    ax.set_xticks(xticks)

    ax.set_yticks(xticks)

    ax.grid()