I want to use sympy.lambdify to compute the derivative of various formulas and evaluate them numerically. Here’s a minimal example:
import sympy as sp
from sympy.utilities.lambdify import lambdify
formula = "arcsin(a**x)"
formula = sp.sympify(formula.strip(), evaluate=False)
d_formula = sp.diff(formula, "x", simplify=False)
grad_func = lambdify(["a","x"], d_formula, modules='numpy')
res = grad_func(0.5,2)
However, when I run this, I get the following error:
---------------------------------------------------------------------------
NameError Traceback (most recent call last)
Cell In[72], line 5
3 d_formula = sp.diff(formula, "x", simplify=False)
4 grad_func = lambdify(["a","x"], d_formula, modules='numpy')
----> 5 grad_func(0.5,2)
File <lambdifygenerated-938>:4, in _lambdifygenerated(a, x)
1 def _lambdifygenerated(a, x):
2 return ( # Not supported in Python with numpy:
3 # Derivative
----> 4 a**x*log(a)*Subs(Derivative(arcsin(_xi_1), _xi_1), xi_1_7206919208181693502, a**x))
NameError: name 'Subs' is not defined
I use .doit() to force evaluation of Subs, but it doesn’t seem to work.
I also change the differentiation options like simplify=False, but it has no effect.
Why does sympy.diff() introduce a Subs expression here and how to avoid it？ How can I transform this derivative into a form that lambdify can handle without errors?
Any insights or alternative approaches would be greatly appreciated!
Using asin and subs:
In [82]: a,x = sp.symbols('a x')
In [83]: formula=sp.asin(a**x); formula
Out[83]:
asin(a**x)
In [84]: d_formula=sp.diff(formula,x, simplify=False); d_formula
Out[84]:
a**x*log(a)/sqrt(1 - a**(2*x))
In [85]: d_formula.subs({a:.5, x:2})
Out[85]:
-0.178969832451953
Now we can use lambdify:
In [86]: foo = sp.lambdify((a,x),d_formula)
In [87]: help(foo)
Help on function _lambdifygenerated:
_lambdifygenerated(a, x)
Created with lambdify. Signature:
func(a, x)
Expression:
a**x*log(a)/sqrt(1 - a**(2*x))
Source code:
def _lambdifygenerated(a, x):
return a**x*log(a)/sqrt(1 - a**(2*x))
In [88]: foo(.5,2)
Out[88]: np.float64(-0.1789698324519529)