Here is my code:
def sigmoid(X, T):
return (1.0 / (1.0 + np.exp(-1.0*np.dot(X, T))))
And this line gives me error
"AttributeError: 'float' object has no attribute 'exp'". X, t are Numpy ndarray.
2 Answers
Probably there's something wrong with the input values for X and/or T. The function from the question works ok:
import numpy as np
from math import e
def sigmoid(X, T):
return 1.0 / (1.0 + np.exp(-1.0 * np.dot(X, T)))
X = np.array([[1, 2, 3], [5, 0, 0]])
T = np.array([[1, 2], [1, 1], [4, 4]])
print(X.dot(T))
# Just to see if values are ok
print([1. / (1. + e ** el) for el in [-5, -10, -15, -16]])
print()
print(sigmoid(X, T))
Result:
[[15 16]
[ 5 10]]
[0.9933071490757153, 0.9999546021312976, 0.999999694097773, 0.9999998874648379]
[[ 0.99999969 0.99999989]
[ 0.99330715 0.9999546 ]]
Probably it's the dtype of your input arrays. Changing X to:
X = np.array([[1, 2, 3], [5, 0, 0]], dtype=object)
Gives:
Traceback (most recent call last):
File "/[...]/stackoverflow_sigmoid.py", line 24, in <module>
print sigmoid(X, T)
File "/[...]/stackoverflow_sigmoid.py", line 14, in sigmoid
return 1.0 / (1.0 + np.exp(-1.0 * np.dot(X, T)))
AttributeError: exp
3 Comments
Il'ya Zhenin
Yes thanks. I didn't know about dtype and used just type(X). I did X = X.astype(float) and it is worked.
gcamargo
so array operations like mean() could not be used in arrays with dype=object? I wonder why?
deeenes
this error message is very misleading: the problem is that the
dtype is in fact numpy.object, but the message says numpy.float64 has no attribute log10, or whatever arithmetic methodYou convert type np.dot(X, T) to float32 like this:
z=np.array(np.dot(X, T),dtype=np.float32)
def sigmoid(X, T):
return (1.0 / (1.0 + np.exp(-z)))
Hopefully it will finally work!
npto a float value.XorThappen to have accidentally been created with a dtype ofobjectinstead offloat64?print X.dtype, T.dtypeand see what you get. You most likely have the wrong dtype.