pydrake.forwarddiff.jacobian 参数的格式？

Question

这个问题最初出现在这里https://github.com/RobotLocomotion/drake/issues/12484。问题是关于如何使用该功能。

功能说明：‖https://drake.mit.edu/pydrake/pydrake.forwarddiff.html#pydrake.forwarddiff.jacobian

我的代码： '''

from pydrake.forwarddiff import jacobian
import math
import numpy as np
def f(t):
return np.array([1.5 - 0.5 * sigmoid(t - 2.5), 0.5, 0.0])
print jacobian(f,0.)

''' 返回的错误信息：

AttributeError Traceback (most recent call last)
in ()
48
49
---> 50 print jacobian(f,0.)
51 #print jacobian(f,np.array[0.])
52

/drake/lib/python2.7/site-packages/pydrake/forwarddiff.py in jacobian(function, x)
38 The function should be vector-input and vector-output.
39 """
---> 40 x_ad = np.empty(x.shape, dtype=np.object)
41 for i in range(x.size):
42 der = np.zeros(x.size)

AttributeError: 'float' object has no attribute 'shape'

If I changed that line into
print jacobian(f,np.array[0.])

then the error becomes:

The autoreload extension is already loaded. To reload it, use:
%reload_ext autoreload

TypeError Traceback (most recent call last)
in ()
49
50 #print jacobian(f,0.)
---> 51 print jacobian(f,np.array[0.])
52
53

TypeError: 'builtin_function_or_method' object has no attribute 'getitem'
If I changed that line into:
print jacobian(f,[0.])
the error:

AttributeError Traceback (most recent call last)
in ()
50 #print jacobian(f,0.)
51 #print jacobian(f,np.array[0.])
---> 52 print jacobian(f,[0.])
53
54 # Run the simulation. Parameters not described above

/drake/lib/python2.7/site-packages/pydrake/forwarddiff.py in jacobian(function, x)
38 The function should be vector-input and vector-output.
39 """
---> 40 x_ad = np.empty(x.shape, dtype=np.object)
41 for i in range(x.size):
42 der = np.zeros(x.size)

AttributeError: 'list' 对象没有属性 'shape'

那么问题来了，我该如何正确使用呢？谢谢

******************更新行************************** ***************

感谢 Eric 的发现。我把代码改成： '''

from pydrake.forwarddiff import jacobian
import math
import numpy as np
def f(t):
    return np.array([1.5 - 0.5 *(t - 2.5)])
def g(t):
    return np.array([1.5 - 0.5 *(t - 2.5), 0.5, 0.0])

x=0.
x = np.asarray(x)
print jacobian(f,x)  # accepted
print jacobian(f,0.) # not accepted
print jacobian(g,x)  # not accepted

''' 结果输出为： '''

[[-0.5]]

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-15-1b94cb728eed> in <module>()
      9 x = np.asarray(x)
     10 print jacobian(f,x)
---> 11 print jacobian(f,0.)

/drake/lib/python2.7/site-packages/pydrake/forwarddiff.py in jacobian(function, x)
     38     The function should be vector-input and vector-output.
     39     """
---> 40     x_ad = np.empty(x.shape, dtype=np.object)
     41     for i in range(x.size):
     42         der = np.zeros(x.size)

AttributeError: 'float' object has no attribute 'shape'

''' 所以，这里至少有两个问题： 1. 每当我想使用 print jacobian(f,0.) 时，我必须先将 float 常量转换为 np.asarray。这很尴尬。 2.注意g(t)的雅可比，该行的错误是： ''' ---------------------------------------------- -------------------------- AttributeError Traceback（最后一次调用）在（） 11 print jacobian(f,x) # 接受 12 #print jacobian(f,0.) # 不被接受 ---> 13 print jacobian(g,x) # 不被接受

/drake/lib/python2.7/site-packages/pydrake/forwarddiff.py in jacobian(function, x)
     45     y_ad = function(x_ad)
     46     return np.vstack(
---> 47         [y.derivatives() for y in y_ad.flat]).reshape(y_ad.shape + (-1,))
     48 
     49 

AttributeError: 'float' object has no attribute 'derivatives'

''' 如果我猜对了，必须将 g(x) 中的所有内容更改为 np.asarray。这很尴尬...

Answer 1

你能保证x是np.ndarray类型吗？为此，请尝试 x = np.asarray(x).

看来我们也可以在 Drake 方面做到这一点，所以我提交了一个 PR（尽管它可能会在新年登陆）： https://github.com/RobotLocomotion/drake/pull/12511

pydrake.forwarddiff.jacobian 参数的格式？

the format of argument of pydrake.forwarddiff.jacobian?

python

jupyter-notebook

drake