precision loss while converting from python float to C++ double

Question

I am embedding python code in my c++ program. The use of PyFloat_AsDouble is causing loss of precision. It keeps only up to 6 precision digits. My program is very sensitive to precision. Is there a known fix for this? Here is the relevant C++ code:

_ret = PyObject_CallObject(pFunc, pArgs);
vector<double> retVals;
for(size_t i=0; i<PyList_Size(_ret); i++){
    retVals[i] = PyFloat_AsDouble(PyList_GetItem(_ret, i));
}

retVals[i] has precision of only 6, while the value returned by the python code is a float that can have a higher precision. How to get full precision?

How is `retVals` declared? If you declared it as an array of `float`, you're getting what you asked for. — Tim Peters, Apr 07 '14 at 22:07

score 2 · Accepted Answer · answered Apr 07 '14 at 22:14

2

Assuming that the Python object contains floating point values stored to double precision, then your code works as you expect.

Most likely you are simply mis-diagnosing a problem that does not exist. My guess is that you are looking at the values in the debugger which only displays the values to a limited precision. Or you are printing them out to a limited precision.

answered Apr 07 '14 at 22:14

David Heffernan

601,492
42
1,072
1,490

Indeed it turns out I wasn't printing with enough precision. I was also misled by this post, which supported my point: https://github.com/coolfluid/coolfluid3/issues/253 – Akhil Apr 07 '14 at 22:25

score -1 · Answer 2 · answered Apr 07 '14 at 22:16

-1

print type(PyList_GetItem(_ret, i))

My bet is it will show float.

Edit: in the python code, not in the C++ code.

answered Apr 07 '14 at 22:16

Fred Mitchell

2,145
2
21
29

Very likely. Of course, Python `float` is almost certainly double precision. – David Heffernan Apr 07 '14 at 22:17

precision loss while converting from python float to C++ double

2 Answers2