NumPy使用手记

来源:互联网 发布:阿里云dns多少 编辑:程序博客网 时间:2024/05/16 06:39

前面一个NumPy系列基本上是抄书,没有多少具体的内容。最近做实验经常使用NumPy,确实感觉到向量计算的强大。这个系列开始,我记录在使用NumPy使用中的一些具体的技巧和注意事项。

1) 巧用 where函数

  where函数是numpy的内置,也是一个非常有用的函数,提供了快速并且灵活的计算功能。

def f_norm_1(data, estimate):
   residule = 0
   for row_index in range(data.shape[0]):
     for column_index in range(data.shape[1]):
       if data[row_index][column_index] != 0:
         residule += (data[row_index][column_index] - estimate[row_index][column_index]) ** 2
   return residule

def f_norm_2(data, estimate) 

    return sum(where(data != 0, (data-estimate) **2, 0))

这两段代码完成同样的功能,计算两个矩阵的差,然后将残差进行平方,注意,因为我需要的是考虑矩阵稀疏性,所以不能用内置的norm,函数1是我用普通的python写的,不太复杂,对于规模10*10的矩阵,计算200次耗时0.15s,函数2使用了where函数和sum函数,这两个函数都是为向量计算优化过的,不仅简介,而且耗时仅0.03s, 快了有五倍,不仅如此,有同学将NumPy和matlab做过比较,NumPy稍快一些,这已经是很让人兴奋的结果。

 

本篇我们看看NumPy中最为基本的Array操作

>>> from numpy import *

创建一个矩阵

>>> a=array([[1,2,3],[4,5,6]])
>>> a.shape
(2, 3)

>>> b=arange(15);print b
[ 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14]
>>> b.reshape(3,5)
array([[ 0,  1,  2,  3,  4],
[ 5,  6,  7,  8,  9],
[10, 11, 12, 13, 14]])

可以看到,A是2行3列的矩阵。通过arange方法,可以得到一个1维的数组。然后我们可以通过reshape方法改变它的维度。

>>> c=zeros((4,5));print c
[[ 0.  0.  0.  0.  0.]
[ 0.  0.  0.  0.  0.]
[ 0.  0.  0.  0.  0.]
[ 0.  0.  0.  0.  0.]]

>>> d=ones((5,7));print d
[[ 1.  1.  1.  1.  1.  1.  1.]
[ 1.  1.  1.  1.  1.  1.  1.]
[ 1.  1.  1.  1.  1.  1.  1.]
[ 1.  1.  1.  1.  1.  1.  1.]
[ 1.  1.  1.  1.  1.  1.  1.]]

>>> e=add(c,arange(20).reshape(4,5))
>>> f=dot(e,d);print f
[[ 10.  10.  10.  10.  10.  10.  10.]
[ 35.  35.  35.  35.  35.  35.  35.]
[ 60.  60.  60.  60.  60.  60.  60.]
[ 85.  85.  85.  85.  85.  85.  85.]]

使用zeros可以生成一个零矩阵。同理,用ones可以生成值全部为1的矩阵。我选择了一个4*5的矩阵e,和一个5*7的矩阵d做点乘。最后得到f矩阵。再举一个更加明显的例子:

>>> a=arange(5);print a
[0 1 2 3 4]
>>> b=arange(5).reshape(5,1);print b
[[0]
[1]
[2]
[3]
[4]]
>>> print dot(a,b)
[30]

点积的效果更加明显了。

ndarray的几个常用属性:

· shape: 代表一个array的形态,是一个向量还是一个矩阵,抑或是一个更复杂的向量组。

· ndim: 代表这个array的维度

· size: 在array中拥有的元素数量

· itemsize: 这个array中每一个元素所需要占的字节数

· nbytes: 这个array的总字节数(=itemsize*size)

· real: 代表一个array中所有元素的实数部分

· imag: 同理,代表一个array中所有元素的虚数部分

· flat: 将这个array整理成一维的,可以索引的一系列的元素组合。它实际上是通过iterator实现的,我们可以通过for x in array.flat来取得到所有的元素

· T: 矩阵转置,同transpose()方法

一些比较有用的方法:

· tolist(): 将array转化成一个Python中的list对象

· item(*args): 取得某一位置的元素

· dump(file): 将这个对象序列化至文件。同cPickle中的dump作用

· dumps(): 将序列化的结果通过字符串加以输出

一些关于Array的形态操作:

· reshape(): 改变array的形态

· resize(): 也是改变array的形态。不同的是,resize是直接修改这个对象的,而reshape则会生成一个新的对象

· transpose(): 这个就是矩阵的转置操作啦

· swapaxes(): 将n个维度中任意两个维度(坐标轴)进行调换

· flatten(): 复制一个一维的array出来

还有一些关于Array的运算操作:

· max():取得所有元素中的最大值

· min():取得最小值。还有一点值得说,就是max、min这些函数都可以针对某一坐标轴(具体维度)进行运算,例如array.max(axis=0),就在0坐标上求最大值

· sum():求和

· cumsum():求累计和

· prod():求所有元素之积

· cumprod():求累计积

· all():如果所有元素都为真,那么返回真;否则返回假

· any():只要有一个元素为真则返回真

· mean():求平均数

 

Array高级操作

1. Vectorize函数

def t(x):  return x + 3a1 = scipy.zeros((5,4))a1NumPy array, format: long[[0 0 0 0] [0 0 0 0] [0 0 0 0] [0 0 0 0] [0 0 0 0]]s = scipy.vectorize(t)a2 = s(a1)a2NumPy array, format: long[[3 3 3 3] [3 3 3 3] [3 3 3 3] [3 3 3 3] [3 3 3 3]]2. NumPy和SciPy相互转化import numpyimport scipya1 = zeros((4,6))type(a1)<type 'scipy.ndarray'>a2 = numpy.asarray(a1)type(a2)<type 'numpy.ndarray'>a3 = numpy.zeros((3,5))type(a3)<type 'numpy.ndarray'>a4 = scipy.asarray(a3)type(a4)<type 'scipy.ndarray'>

 

 

NumPy 数学函数

Trigonometric functions



sin (x[, out])Trigonometric sine, element-wise.cos (x[, out])Cosine elementwise.tan (x[, out])Compute tangent element-wise.arcsin (x[, out])Inverse sine elementwise.arccos (x[, out])Trigonometric inverse cosine, element-wise.arctan (x[, out])Trigonometric inverse tangent, element-wise.hypot (x1, x2[, out])Given two sides of a right triangle, return its hypotenuse.arctan2 (x1, x2[, out])Elementwise arc tangent of x1/x2 choosing the quadrant correctly.degrees (x[, out])Convert angles from radians to degrees. This is the same function as rad2deg but the latter is preferred because of the more descriptive name.radians (x[, out])Convert angles from degrees to radians. This function is the same as deg2rad, which is more descriptive..unwrap (p[, discont, axis])Unwrap by changing deltas between values to 2*pi complement.

Hyperbolic functions



sinh (x[, out])Hyperbolic sine, element-wise.cosh (x[, out])Hyperbolic cosine, element-wise.tanh (x[, out])Hyperbolic tangent element-wise.arcsinh (x[, out])Inverse hyperbolic sine elementwise.arccosh (x[, out])Inverse hyperbolic cosine, elementwise.arctanh (x[, out])Inverse hyperbolic tangent elementwise.

Rounding



around (a[, decimals, out])Evenly round to the given number of decimals.round_ (a[, decimals, out])Round an array to the given number of decimals.rint (x[, out])Round elements of the array to the nearest integer.fix (x[, y])Round to nearest integer towards zero.floor (x[, out])Return the floor of the input, element-wise.ceil (x[, out])Return the ceiling of the input, element-wise.

Sums, products, differences



prod (a[, axis, dtype, out])Return the product of array elements over a given axis.sum (a[, axis, dtype, out])Return the sum of array elements over a given axis.nansum (a[, axis])Return the sum of array elements over a given axis treating Not a Numbers (NaNs) as zero.cumprod (a[, axis, dtype, out])Return the cumulative product of elements along a given axis.cumsum (a[, axis, dtype, out])Return the cumulative sum of the elements along a given axis.diff (a[, n, axis])Calculate the nth order discrete difference along given axis.ediff1d (ary[, to_end, to_begin])The differences between consecutive elements of an array.gradient (f, *varargs)Return the gradient of an N-dimensional array.cross (a, b[, axisa, axisb, axisc, ...])Return the cross product of two (arrays of) vectors.trapz (y[, x, dx, axis])Integrate along the given axis using the composite trapezoidal rule.

Exponents and logarithms



exp (x[, out])Calculate the exponential of the elements in the input array.expm1 (x[, out])Return the exponential of the elements in the array minus one.log (x[, out])Natural logarithm, element-wise.log10 (x[, out])Compute the logarithm in base 10 element-wise.log2 (x[, y])Return the base 2 logarithm.log1p (x[, out])log(1 + x) in base e, elementwise.

Other special functions



i0 (x)Modified Bessel function of the first kind, order 0.sinc (x)Return the sinc function.

Floating point routines



signbit (x[, out])Returns element-wise True where signbit is set (less than zero).frexp (x[, out1, out2])Split the number, x, into a normalized fraction (y1) and exponent (y2)ldexp (x1, x2[, out])Compute y = x1 * 2**x2.

Arithmetic operations



add (x1, x2[, out])Add arguments element-wise.reciprocal (x[, out])Return element-wise reciprocal.negative (x[, out])Returns an array with the negative of each element of the original array.multiply (x1, x2[, out])Multiply arguments elementwise.divide (x1, x2[, out])Divide arguments element-wise.power (x1, x2[, out])Returns element-wise base array raised to power from second array.subtract (x1, x2[, out])Subtract arguments element-wise.true_divide (x1, x2[, out])Returns an element-wise, true division of the inputs.floor_divide (x1, x2[, out])Return the largest integer smaller or equal to the division of the inputs.fmod (x1, x2[, out])Return the remainder of division.mod (x1, x2[, out])Returns element-wise remainder of division.modf (x[, out1, out2])Return the fractional and integral part of a number.remainder (x1, x2[, out])Returns element-wise remainder of division.

Handling complex numbers



angle (z[, deg])Return the angle of the complex argument.real (val)Return the real part of the elements of the array.imag (val)Return the imaginary part of array.conj (x[, out])Return the complex conjugate, element-wise.

Miscellaneous



convolve (a, v[, mode])Returns the discrete, linear convolution of two one-dimensional sequences.clip (a, a_min, a_max[, out])Clip (limit) the values in an array.sqrt (x[, out])Return the positive square-root of an array, element-wise.square (x[, out])Return the element-wise square of the input.absolute (x[, out])Calculate the absolute value element-wise.fabs (x[, out])Compute the absolute values elementwise.sign (x[, out])Returns an element-wise indication of the sign of a number.maximum (x1, x2[, out])Element-wise maximum of array elements.minimum (x1, x2[, out])Element-wise minimum of array elements.nan_to_num (x)Replace nan with zero and inf with large numbers.real_if_close (a[, tol])If complex input returns a real array if complex parts are close to zero.interp (x, xp, fp[, left, right])One-dimensional linear interpolation.

0 0