Numpy的统计分析
其他
2018-06-05 06:05:45
阅读次数: 1
'''----------------------保存数据savetxt-----------------------'''import numpy as npexample = np.arange(24).reshape(3,8)'''将numpy的ndarray数组保存到文本文件中：savetxt(fileName,data)fileName:保存文件路径和名称data:需要保存的数据'''np.savetxt("./data/example.txt",example)print("保存完成~")
'''---------------------加载数据loadtxt-----------------------'''import numpy as np'''numpy加载csv文件数据np.loadtxt(filepath,delimiter,usecols,converters,unpack)filepath:加载文件的路径delimiter:数据文件的分隔符usecols：加载数据文件中数据列索引unpack:当加载多列数据时是否需要将数据列进行解耦赋值给不同的变量converters:{}，第一个参数为要加载的数据是第几列，第二个参数是对数据操作的函数或表达式'''result = np.loadtxt('data/data.csv',delimiter=",",usecols=(6,7))print(result)#从result中切片分别取出收盘价和成交量close = result[:,0]print("收盘价:\n",close)amount = result[:,1]print("成交量:\n",amount)#当加载csv文件的多列数据时可以使用unpack将加载的数据列进行解耦到不同的数组中arr1,arr2 = np.loadtxt('data/data.csv',delimiter=",",usecols=(6,7),unpack=True)print("收盘价:\n",arr1)print("成交量:\n",arr2)#加载收盘价并将数据转化为整数close = np.loadtxt("data/data.csv",delimiter=",",usecols=(6), converters={6:lambda x: int(float(x))})print(close)
'''---------------------数组的属性-----------------------'''import numpy as npexample = np.arange(24).reshape(3,4,2)#01->ndim数组的维度print("01维度：",example.ndim) #3#02->size数组元素的个数print("02元素个数：",example.size) #24#03->lenprint("03元素个数：",len(example)) #3#04->itemsize数组中的元素在内存中所占的字节数print("每个元素在内存中所占的字节数：",example.itemsize) #int32/8=4#05->nbytes整个数组所占的存储空间/字节数print("整个数组所占的存储空间/字节数:",example.nbytes) #24*4=96print("整个数组所占的存储空间/字节数:",example.size*example.itemsize) #24*4=96#06->T数组的转置结果print("原数组:\n",example) #3x4x2print("转置结果:\n",example.T) #2x4x3#07->real复数数组的实部comp = np.array([2+5j,6+3j,4+2j])real = comp.realprint("复数数组的实部是:\n",real) #[2. 6. 4.]#08->imag复数数组的虚部imag = comp.imagprint("复数数组的虚部是：",imag) # [5. 3. 2.]#09求comp复数数组的每个元素的模print("复数数组的每个元素的模是:",np.sqrt(real**2+imag**2)) #[5.38516481 6.70820393 4.47213595]iterator = np.arange(9).reshape(3,3)#10将数组展平成扁平迭代器result = iterator.flatfor item in result: print(item) #0 1 2 3 4 5 6 7 8
'''---------------数组的分割split/hsplit/vsplit/dsplit------------'''import numpy as np#数组的分割full = np.arange(9).reshape(3,3)print(full)#np.split 当axis=0则是按行进行分割axis0 = np.split(full,3,axis=0)print(axis0[0]) #[[0 1 2]]print(axis0[0].shape) #(1,3)print(axis0[1]) #[[3 4 5]]print(axis0[2]) #[[6 7 8]]#np.split 当axis=1则是按列进行分割axis1 = np.split(full,3,axis=1)print(axis1)#hsplit(arg,num),与np.split当axis=1一样水平分割与按列分割一样axis2 = np.hsplit(full,3)print(axis2)#vsplit(arg,num),与np.split当axis=0一样垂直分割与按行分割一样axis3 = np.vsplit(full,3)print(axis3)#2x2x2d3 = np.arange(8).reshape(2,2,2)print("d3:\n",d3)#dsplit深度分割,dsplit only works on arrays of 3 or more dimensionsdsplit = np.dsplit(d3,2)print(dsplit)print(dsplit[0]) #2x2x1print(dsplit[1]) #2x2x1#2x2x6d4 = np.arange(24).reshape(2,2,6)print("d4:\n",d4)#dsplit深度分割,dsplit only works on arrays of 3 or more dimensionsdsplit = np.dsplit(d4,2)print(dsplit)print(dsplit[0]) #2x2x1print(dsplit[1]) #2x2x1
'''---------------股票波动率------------'''import numpy as npclose = np.loadtxt("data/data.csv",delimiter=",",usecols=(6))#计算简单的收益率：相邻两天的差除以前一天的价格returns = np.diff(close)/close[:-1]print(returns)#对数收益率#对所有收盘价取对数logClose = np.log(close)print("收盘价的对数：",logClose)diffLogClose = np.diff(logClose)print("收盘价的对数收益率：",diffLogClose)#对数收益率的标准差std = np.std(diffLogClose)print("对数收益率的标准差",std)#计算年波动率annual_volatility = std/np.mean(diffLogClose)year_volatility = annual_volatility/np.sqrt(1/252)print("年波动率是：",year_volatility)#计算月波动率month_volatility = annual_volatility/np.sqrt(1/12)print("月波动率是：",month_volatility)
'''---------------统计方法average/mean/max/min------------'''import numpy as np#加载收盘价和成交量close,amount = np.loadtxt("data/data.csv",delimiter=",", usecols=(6,7),unpack=True)print("收盘价:\n",close)print("成交量:\n",amount)#01->成交量加权平均价:np.average(ndarray,weights)aver = np.average(close,weights=amount)print("成交量加权平均价:",aver)#02->收盘价算数平均值:np.mean(ndarray) np.average(ndarray)mean = np.mean(close)average = np.average(close) #350.5895493532009print("收盘价算数平均值:",mean) #351.0376666666667print("收盘价算数平均值:",average) #351.0376666666667#03->求最大值最小值 np.max(ndarray) np.min(ndarray)#求成交量最大值和最小值maxAmount = np.max(amount)print("最大成交量：",maxAmount)minAmount = np.min(amount)print("最小成交量：",minAmount)example = np.arange(24).reshape(2,3,4)print(example)print(np.max(example))print(np.min(example))#自定义一个函数获取数组的最大值最小值def getMaxandMin(ndarray): max,min = ndarray[0],ndarray[0] result = ndarray.flat #先展平 for item in result: if item>max: max = item if item<min: min = item return max,minmax,min = getMaxandMin(example)print(max,min)
'''---------------统计方法ptp/median/var/repeat---------------'''import numpy as np#04->计算数组的极差(最大值和最小值的差值 ) np.ptphigh = np.loadtxt("data/data.csv",delimiter=",",usecols=(4))print(high)#计算最高价的极差ptp = np.ptp(high)print("最高价的极差",ptp)#利用最大值最小值得到极差max = np.max(high)min = np.min(high)ptp2 = max - minprint("ptp2：",ptp2)#05->计算收盘价的中位数 np.median()close = np.loadtxt("data/data.csv",delimiter=",",usecols=(6))print(close)median = np.median(close)print("收盘价的中位数:",median)list(close)# print(close)close.sort()print(close)if close.size%2 !=0: median1 = close[(close.size-1)/2]else: median1 = (close[int(close.size/2)]+close[int(close.size/2)-1])/2print(median1)#06->计算数据样本的方差 np.var()print("收盘价的方差：",np.var(close))open = np.loadtxt("data/data.csv",delimiter=",",usecols=(3))print("开盘价的方差：",np.var(open))#07->np.repeat()a = np.array([1,3,5])times = np.array([4,6,2])b = np.repeat(a,times)print(b) #[1,1,1,1,3,3,3,3,3,3,5,5]
'''---------------统计方法prod/cumprod/compress---------------'''import numpy as np#1.求阶乘 np.prodarr = np.array([5,2,7,4])prod = np.prod(arr)print("阶乘：",prod) # 5*2*7*4=280#2.累积乘积 np.cumprod()cumArr = np.array([2,3,4,5])cumprod = np.cumprod(cumArr)print(cumprod) #[ 2 6 24 120]#3.将数组中所有比num1小的元素设为num1，# 所有比num2大的元素设为num2，# 介于num1和num2之间的数值保持不变 np.clip(num1,num2)test = np.array([3,56,7,44,22,18])print("clip",test.clip(48,20)) #clip [48 20 48 48 48 48]与num1比小，与num2比大#4.np.compress()返回根据一定条件筛选后的数组close = np.loadtxt("data/data.csv",delimiter=",",usecols=(6))#计算对数收益率logReturns = np.diff(np.log(close))print("对数收益率：",logReturns)#筛选出对数收益率中大于0 的元素mask = logReturns>0print("mask索引：",mask)result= logReturns.compress(mask)print("对数收益率中大于0 的元素:",result)
'''---------------数组数据的筛选---------------'''import numpy as npclose = np.loadtxt("data/data.csv",delimiter=",",usecols=(6))#对数收益率logReturns = np.diff(np.log(close))print(logReturns)#筛选出对数收益率大于0的所有元素的索引indices= np.where(logReturns>0)print("对数收益率大于0的索引为：",indices)#根据索引值，获取相应的数组元素result = np.take(logReturns,indices)print(result)
'''---------------日期格式转化---------------'''from datetime import datetimedateStr = "2018-5-28"newdate = datetime.strptime(dateStr,"%Y-%m-%d")print(newdate) #2018-05-28 00:00:00#获取dateStr日期对应的星期weekday = newdate.date().weekday()print(type(weekday))print("{}是星期{}".format(dateStr,weekday+1)) #2018-5-28是星期1
'''---------------案例---------------'''import numpy as npfrom datetime import datetime#将日期格式字符串转化得到星期数def datestr2num(s): print(type(s)) # 加载过来的第一列日期数据是bytes类型，需转化为str return datetime.strptime(s.decode("utf8"),"%d-%m-%Y").date().weekday()#converters参数：“1”表示要转化的是第一列数据，datestr2num是指使用哪个函数做转化dates,close = np.loadtxt("data/data.csv",delimiter=",",usecols=(1,6), converters={1:datestr2num},unpack=True)print("日期：",dates)print("收盘价：",close)aver = []for i in range(5): #根据星期数条件得到一个mask数组 mask = dates==i #根据mask数组筛选出星期数为i的所有的收盘价 weekday = close.compress(mask) #计算出星期数为i的所有收盘价的均值 dayAver = np.mean((weekday)) #将星期数为i的收盘价的均值添加到aver数组中 aver.append(dayAver) #aver里存的是周一到周五各个的平均值print(aver)#获取数组中最大值的索引maxIndex = np.argmax(aver)minIndex = np.argmin(aver)print("星期{}的平均收盘价最高".format(maxIndex+1))print("星期{}的平均收盘价最低".format(minIndex+1))
转载自blog.csdn.net/qq_42379006/article/details/80562699