C# and .NET examples · Python and .NET integration

Call Python from C# with Python.NET

A .NET 8 console application demonstrating how to call Python code from C# using Python.NET (pythonnet 3.0.5) — calling Python functions, passing C# objects to Python, running a scikit-learn model, and hosting a FastAPI app from a .NET process. The project covers four integration scenarios — from basic arithmetic to ML model inference and launching a FastAPI server.

In this guide: Get started · Walk through the example · Understand the details · Check the limits

🔎 Overview

Python.NET embeds a Python interpreter (CPython) inside the .NET process. The Py.GIL() block acquires the Global Interpreter Lock, after which you can import Python modules, call functions, and exchange objects between C# and Python dynamically.

🚀 Quick start

Build and Run

After installing the matching Python DLL and packages, build and run from the output directory so model training and loading use the same location:

cd PythonBridge
dotnet build
cd bin/Debug/net8.0
dotnet PythonBridge.dll

Program.cs invokes the four scenarios sequentially and waits for input if they all return. Earlier embedding or model-loading failures can prevent later scenarios from running. The PyScript/*.py files are copied to the build output, and each caller adds that folder to Python’s sys.path before importing the modules.

📦 Prerequisites and installation

Requirements

• .NET 8 SDK
• Python 3.12 installed with python312.dll accessible (system PATH or application directory). Each caller sets Runtime.PythonDLL = "python312.dll"; change it to match your Python version.
• Python packages:

python -m pip install joblib numpy scikit-learn fastapi uvicorn

NuGet Packages

Package	Version
pythonnet	3.0.5

💡 Usage

Scenarios

1. Basic Operations (Basic.Execute)

Imports basic_operations.py and calls:

Python Function	Description
add_numbers(10, 20)	Returns the sum
multiply_numbers(5, 6)	Returns the product
get_greeting("John")	Returns a greeting string

Also reads the module-level constant PI.

2. Log Formatting (LogFormat.Execute)

Imports logger.py inside a named Python scope. Passes C# LogEntry objects (Level, Message) to logger.format_entry(entry) and prints the formatted output.

3. Iris ML Inference (IrisCode.Execute)

• If iris_nb.pkl does not exist, runs train_iris_model.py to train a Naive Bayes classifier on the Iris dataset and save the model
• Loads the saved model with joblib, builds a NumPy input array, and prints the predicted class for [5.1, 3.5, 1.4, 0.2]
• Uses PythonEngine.BeginAllowThreads / EndAllowThreads to release the GIL during non-Python work

4. FastAPI Server (ApiCode.Execute)

Imports api.py (which defines a FastAPI app) and starts it with uvicorn on 127.0.0.1:8000. The app exposes two GET endpoints:

Endpoint	Response
/	{"message": "Hello, World! from Python.NET with FastAPI."}
/hello/{name}	{"message": "Hello, {name}! from Python.NET with FastAPI."}

uvicorn.run blocks while the server is running, so the application stays on this step until you stop the server (for example with Ctrl+C).

Python Scripts (PyScript/)

Script	Description
basic_operations.py	add_numbers, multiply_numbers, get_greeting, and PI constant
logger.py	format_entry(entry) — formats a log entry object
train_iris_model.py	Trains a Naive Bayes classifier on Iris data, saves iris_nb.pkl
api.py	FastAPI application

🏗️ Architecture

Project Structure

PythonBridge/
├── PyScript/
│   ├── basic_operations.py      # Arithmetic and greeting functions
│   ├── logger.py                # Log entry formatter
│   ├── train_iris_model.py      # Iris model training script
│   └── api.py                   # FastAPI application
├── PythonCaller/
│   ├── Basic.cs                 # Calls basic_operations.py
│   ├── LogFormat.cs             # Calls logger.py with LogEntry objects
│   ├── IrisCode.cs              # Trains or loads Iris model, runs inference
│   └── ApiCode.cs               # Starts FastAPI server via uvicorn
├── LogEntry.cs                  # Log entry DTO (Level, Message)
├── Program.cs                   # Entry point — runs all four scenarios
├── PythonBridge.csproj
└── PythonBridge.sln

🧪 Development and testing

From the repository root, dotnet build PythonBridge/PythonBridge.csproj checks the .NET project. Python imports, embedding, model training and the FastAPI server need separate runtime checks; none were launched during this documentation review.

🛠️ Troubleshooting and limitations

Troubleshooting

• Runtime.PythonDLL / DllNotFoundException — Python.NET cannot find the Python library. Install the matching Python version, or set Runtime.PythonDLL to the full path of your pythonXY.dll.
• ModuleNotFoundError — install the packages listed above into the same Python installation that the DLL belongs to.
• Port 8000 already in use — stop the other process or change the port in ApiCode.cs.

The trainer writes iris_nb.pkl relative to the current working directory, while IrisCode looks for it in AppDomain.CurrentDomain.BaseDirectory. With dotnet run from the source folder these locations usually differ, so first-run training can succeed but loading can fail before FastAPI starts. Build first, then run the DLL from its output directory to align the paths, or review a code fix separately. Each scenario initializes and shuts down Python; repeated interpreter initialization and native package compatibility were not runtime-tested. Only load trusted joblib/pickle models.

🔗 Related documentation

Related Examples

• The reverse direction: call a C# library from Python
• Back to all C# and .NET code examples

---

Documentation links

Read all C# and .NET Code Examples guides online · Project overview · All developer projects