This is the home of an executable book project about using Modern C++ for high-performance data science.
It’s a companion to a series of talks by Armin Sobhani for the Compute Ontario Colloquia.
It’ll be updated as more talks in the series are delivered.
Try in a Container 🛠️
docker run -p 8888:8888 -it --rm asobhani/high-performance-data-science-with-modern-cppOr this one with CUDA support:
docker run --gpus=all -p 8888:8888 -it --rm asobhani/high-performance-data-science-with-modern-cpp:latest-cudaapptainer run docker://asobhani/high-performance-data-science-with-modern-cpp:latestOr this one with CUDA support:
apptainer run --nv docker://asobhani/high-performance-data-science-with-modern-cpp:latest-cudaTry on Binder 🛠️
👉
Be advised sometimes it takes several minutes to start!
Watch the Recordings 📺¶
1️⃣ Xeus-Cling and G3P
2️⃣ Ranx
C++ vs. Python for Data Science¶
😌 Ease of Use
📚 Community and Libraries
Winner 🏆 Python 🏆 | |
|---|---|
| 🔻 | C++'s ecosystem is not as extensive as Python’s for data science |
| ✅ | Python has extensive libraries like NumPy, Pandas, Matplotlib, etc. and a large and active community |
🏃 Performance
🔀 Concurrency
Winner 🏆 C++ 🏆 | |
|---|---|
| ✅ | C++ has built-in support for concurrency (C++11) and parallel algorithms (C++17) |
| 🔻 | Python’s global interpreter lock can be a limitation for multi-threaded applications |
💼 Memory Management
💫 Rapid Prototyping
Winner 🏆 Python 🏆 | |
|---|---|
| 🔻 | C++'s compiled nature makes it a lackluster |
| ✅ | Python’s interpreted nature combined with Project Jupyter makes it a perfect match for the job |