在C++中嵌入Python：无法将字符串或char*转换为numpy数组

Hey there! Let's figure out how to turn your C++ string/char* into a numpy array when embedding Python. I’ve got two solid approaches for you—pick the one that fits your workflow best:

If you want minimal changes to your existing C++ code, let Python do the string-to-numpy conversion. Just modify your Python function to split the input string, convert it to numerical values, and wrap it in a numpy array.

Example Python Function #

import numpy as np

def filter_func(input_str):
    # Split the comma-separated string into individual number strings
    num_strings = input_str.split(',')
    # Convert to floats and create a numpy array
    arr = np.array([float(s) for s in num_strings])
    # Add your processing logic here...
    return arr

Your C++ Code (Almost Unchanged) #

Keep your existing C++ code that passes the string to Python—now when you call filterFunc, the input gets converted to a numpy array inside Python automatically:

const char* blu = "1,2,3,4,5,6,7,8,9";
PyObject* args = Py_BuildValue("(s)", blu);

if (!args) {
    PyErr_Print();
    Py_DECREF(filterFunc);
    qDebug()<<"Error building args tuple";
    return;
}

PyObject* resultObj = PyObject_CallObject(filterFunc, args);

// Clean up references
Py_DECREF(args);

This is great for quick prototyping or if you don’t need to manipulate the numpy array directly in C++.

For better performance (especially with large datasets) or if you need full control over the array creation in C++, use numpy’s C API to build the array directly.

Step 1: Set Up Numpy C API #

First, include the numpy headers and initialize the API (must be done after Py_Initialize()):

#include <numpy/arrayobject.h>
#include <sstream>
#include <vector>

// Initialize numpy's C API (call this once after Py_Initialize())
import_array();

Step 2: Parse String to Numerical Data & Build Numpy Array #

const char* blu = "1,2,3,4,5,6,7,8,9";

// Parse the comma-separated string into a vector of floats
std::vector<float> values;
std::stringstream ss(blu);
std::string token;

while (std::getline(ss, token, ',')) {
    try {
        values.push_back(std::stof(token));
    } catch (const std::exception& e) {
        qDebug() << "Failed to parse number:" << token.c_str();
        // Handle error appropriately
        return;
    }
}

// Define the array shape (1D array with length equal to values.size())
npy_intp dims[1] = {static_cast<npy_intp>(values.size())};

// Create the numpy array from our C++ data buffer
PyObject* numpyArray = PyArray_SimpleNewFromData(
    1,                  // Number of dimensions
    dims,               // Array shape
    NPY_FLOAT,          // Data type (use NPY_DOUBLE for double-precision)
    values.data()       // Pointer to our numerical data
);

// Critical: Tell numpy we own the data buffer (so it doesn't try to free it)
// If you want numpy to take ownership of the data, use PyArray_SetBaseObject with a PyObject wrapper
PyArray_SetBaseObject((PyArrayObject*)numpyArray, NULL);

// Pass the numpy array to your Python function
PyObject* args = Py_BuildValue("(O)", numpyArray);
if (!args) {
    PyErr_Print();
    Py_DECREF(filterFunc);
    Py_DECREF(numpyArray);
    qDebug()<<"Error building args tuple";
    return;
}

PyObject* resultObj = PyObject_CallObject(filterFunc, args);

// Clean up all references to avoid memory leaks
Py_DECREF(args);
Py_DECREF(numpyArray);

Key Notes for This Approach: #

• Linking Numpy: When compiling, make sure to link against numpy’s libraries (use pkg-config --libs numpy or specify the library path manually).
• Memory Management: PyArray_SimpleNewFromData doesn’t take ownership of your data buffer by default—you’re responsible for keeping values alive until numpyArray is no longer used. If you want numpy to handle memory cleanup, wrap your data in a PyBytes object and use PyArray_SetBaseObject.
• Error Checking: Always check if numpy API calls return NULL—use PyErr_Print() to debug any issues.

内容的提问来源于stack exchange，提问作者Alok