一维插值
插值不同于拟合。插值函数经过样本点,拟合函数一般基于最小二乘法尽量靠近所有样本点穿过。常见插值方法有拉格朗日插值法、分段插值法、样条插值法。
拉格朗日插值多项式:当节点数n较大时,拉格朗日插值多项式的次数较高,可能出现不一致的收敛情况,而且计算复杂。随着样点增加,高次插值会带来误差的震动现象称为龙格现象。
分段插值:虽然收敛,但光滑性较差。
样条插值:样条插值是使用一种名为样条的特殊分段多项式进行插值的形式。由于样条插值可以使用低阶多项式样条实现较小的插值误差,这样就避免了使用高阶多项式所出现的龙格现象,所以样条插值得到了流行。
-
-
- import numpy as np
- from scipy import interpolate
- import pylab as pl
-
- x=np.linspace(0,10,11)
-
- y=np.sin(x)
- xnew=np.linspace(0,10,101)
- pl.plot(x,y,"ro")
-
- for kind in ["nearest","zero","slinear","quadratic","cubic"]:
-
-
-
- f=interpolate.interp1d(x,y,kind=kind)
-
- ynew=f(xnew)
- pl.plot(xnew,ynew,label=str(kind))
- pl.legend(loc="lower right")
- pl.show()
结果:
二维插值
方法与一维数据插值类似,为二维样条插值。
-
-
-
-
- import numpy as np
- from scipy import interpolate
- import pylab as pl
- import matplotlib as mpl
-
- def func(x, y):
- return (x+y)*np.exp(-5.0*(x**2 + y**2))
-
-
- y,x= np.mgrid[-1:1:15j, -1:1:15j]
-
- fvals = func(x,y)
- print len(fvals[0])
-
-
- newfunc = interpolate.interp2d(x, y, fvals, kind='cubic')
-
-
- xnew = np.linspace(-1,1,100)
- ynew = np.linspace(-1,1,100)
- fnew = newfunc(xnew, ynew)
-
-
-
-
- pl.subplot(121)
- im1=pl.imshow(fvals, extent=[-1,1,-1,1], cmap=mpl.cm.hot, interpolation='nearest', origin="lower")
-
- pl.colorbar(im1)
-
- pl.subplot(122)
- im2=pl.imshow(fnew, extent=[-1,1,-1,1], cmap=mpl.cm.hot, interpolation='nearest', origin="lower")
- pl.colorbar(im2)
-
- pl.show()
左图为原始数据,右图为二维插值结果图。
二维插值的三维展示方法
-
-
-
-
-
- import numpy as np
- from mpl_toolkits.mplot3d import Axes3D
- import matplotlib as mpl
- from scipy import interpolate
- import matplotlib.cm as cm
- import matplotlib.pyplot as plt
-
- def func(x, y):
- return (x+y)*np.exp(-5.0*(x**2 + y**2))
-
-
- x = np.linspace(-1, 1, 20)
- y = np.linspace(-1,1,20)
- x, y = np.meshgrid(x, y)
-
- fvals = func(x,y)
-
- fig = plt.figure(figsize=(9, 6))
-
- ax=plt.subplot(1, 2, 1,projection = '3d')
- surf = ax.plot_surface(x, y, fvals, rstride=2, cstride=2, cmap=cm.coolwarm,linewidth=0.5, antialiased=True)
- ax.set_xlabel('x')
- ax.set_ylabel('y')
- ax.set_zlabel('f(x, y)')
- plt.colorbar(surf, shrink=0.5, aspect=5)
-
-
- newfunc = interpolate.interp2d(x, y, fvals, kind='cubic')
-
-
- xnew = np.linspace(-1,1,100)
- ynew = np.linspace(-1,1,100)
- fnew = newfunc(xnew, ynew)
- xnew, ynew = np.meshgrid(xnew, ynew)
- ax2=plt.subplot(1, 2, 2,projection = '3d')
- surf2 = ax2.plot_surface(xnew, ynew, fnew, rstride=2, cstride=2, cmap=cm.coolwarm,linewidth=0.5, antialiased=True)
- ax2.set_xlabel('xnew')
- ax2.set_ylabel('ynew')
- ax2.set_zlabel('fnew(x, y)')
- plt.colorbar(surf2, shrink=0.5, aspect=5)
-
- plt.show()
左图的二维数据集的函数值由于样本较少,会显得粗糙。而右图对二维样本数据进行三次样条插值,拟合得到更多数据点的样本值,绘图后图像明显光滑多了。
参考链接:
1.拉格朗日插值法:https://zh.wikipedia.org/wiki/拉格朗日插值法
2.样条插值:https://zh.wikipedia.org/wiki/样条插值