在Visual Studio 2017中从C#向Python传递参数（用于gensim）

Hey there! As a fellow C# dev who's made the jump to Python for NLP tasks, I totally get how overwhelming this interop stuff can feel at first. Let's break this down step by step to get your data flowing between C# and Python smoothly in Visual Studio 2017.

First, make sure you've got Python Tools for Visual Studio (PTVS) set up—it's included with the Python development workload in VS2017, but if you skipped that during installation, you can add it via the VS Installer.

You’ve got two solid options for passing data between the languages:

• Embed Python directly in your C# app using Python.NET (best for tight, frequent integration)
• Serialize data to JSON and pass it between separate C# and Python processes (simpler for one-off calls)

First, a quick reminder: dictionary.doc2bow(doc) returns a list of tuples like [(token_id_1, count_1), (token_id_2, count_2), ...], where the first value is an integer token ID and the second is a numeric weight/count. Your C# List<List<double>> needs to map directly to this structure—each inner list should hold two values: the token ID (which we’ll cast to int) and the count/weight.

Option 1: Python.NET for Tight Integration #

This lets you call Python functions directly from your C# code, which is perfect if you want to keep everything within a single VS solution.

Step 1: Install Python.NET #

• In your C# project, install the Python.Runtime NuGet package.
• Double-check that your VS Python environment matches the one you’re using for NLP work (go to Tools > Python > Python Environments to confirm paths and packages align).

Step 2: Convert & Pass the C# List to Python #

Here’s a working code snippet to handle the conversion and interop:

using Python.Runtime;
using System.Collections.Generic;

class Program
{
    static void Main(string[] args)
    {
        // Initialize the Python runtime
        PythonEngine.Initialize();

        // Your sample C# data: List<List<double>>
        var csharpTermMatrix = new List<List<double>>()
        {
            new List<double> {0, 2.5},
            new List<double> {1, 1.0},
            new List<double> {3, 4.2}
        };

        // Acquire Python's Global Interpreter Lock (GIL) before interacting with Python objects
        using (Py.GIL())
        {
            var pyTermMatrix = new PyList();
            foreach (var entry in csharpTermMatrix)
            {
                // Convert each inner list to a Python tuple (doc2bow expects this format)
                // Cast the first value to int since token IDs are integers
                var tokenId = (int)entry[0];
                var count = entry[1];
                var pyTuple = new PyTuple(new PyObject[] { tokenId.ToPython(), count.ToPython() });
                pyTermMatrix.Append(pyTuple);
            }

            // Import your Python NLP script and pass the converted data
            dynamic sys = Py.Import("sys");
            sys.path.Append(@"C:\path\to\your\python\scripts"); // Update this path!
            dynamic nlpScript = Py.Import("your_nlp_module");
            nlpScript.process_term_matrix(pyTermMatrix); // Call your Python function
        }

        // Clean up the Python runtime
        PythonEngine.Shutdown();
    }
}

Step 3: Update Your Python Script #

Modify your Python script to accept the passed data and replace line 48’s doc_term_matrix:

def process_term_matrix(doc_term_matrix):
    # Replace line 48's hardcoded doc_term_matrix with this parameter
    # Example usage with Gensim:
    from gensim.models import LdaModel
    lda_model = LdaModel([doc_term_matrix], num_topics=5)
    # Rest of your NLP logic here...

Option 2: JSON for Simple Cross-Process Data Passing #

If you prefer to keep your C# and Python code as separate processes, JSON serialization is a low-fuss approach.

Step 1: Serialize C# Data to JSON #

Use System.Text.Json (or Newtonsoft.Json) to turn your list into a JSON string:

using System.Text.Json;
using System.Collections.Generic;
using System.Diagnostics;

class Program
{
    static void Main(string[] args)
    {
        var csharpTermMatrix = new List<List<double>>()
        {
            new List<double> {0, 2.5},
            new List<double> {1, 1.0},
            new List<double> {3, 4.2}
        };

        // Serialize the list to JSON
        string jsonData = JsonSerializer.Serialize(csharpTermMatrix);

        // Launch your Python script and pass the JSON as a command-line argument
        var processStartInfo = new ProcessStartInfo
        {
            FileName = @"C:\path\to\your\python.exe", // Update this path!
            Arguments = $"your_nlp_script.py \"{jsonData}\"",
            RedirectStandardOutput = true,
            UseShellExecute = false,
            CreateNoWindow = true
        };

        using (var process = Process.Start(processStartInfo))
        {
            string pythonOutput = process.StandardOutput.ReadToEnd();
            process.WaitForExit();
            // Handle output from Python if needed
        }
    }
}

Step 2: Deserialize & Convert in Python #

Parse the JSON and convert it to the doc2bow-compatible tuple format:

import json
import sys

def main():
    # Grab the JSON data from the command line
    json_input = sys.argv[1]
    csharp_list = json.loads(json_input)
    
    # Convert to the list-of-tuples format doc2bow uses
    doc_term_matrix = [(int(token_id), count) for token_id, count in csharp_list]
    
    # Replace line 48's doc_term_matrix with this variable
    # Rest of your NLP code...

if __name__ == "__main__":
    main()

• Always use the GIL with Python.NET: Python requires the Global Interpreter Lock for thread safety—wrap all Python interactions in using (Py.GIL()).
• Token IDs must be integers: doc2bow expects token IDs to be whole numbers, so make sure you cast the first element of each inner C# list to int before converting to Python.
• Match Python environments: Ensure your C# project uses the same Python installation as your NLP scripts (check VS’s Python Environments settings to confirm).

内容的提问来源于stack exchange，提问作者AyaZoghby