Computational & Deep Learning Frameworks Benchmark on real-world tasks

Use benchmark.ipynb to run the benchmark.

Frameworks compared:

• Python 3/Numpy with or without Numba JIT compilation
• Julia The Julia Programming Language
• C++ called from Python using pybind11
• Tensorflow v1 and v2 in CPU/GPU mode
• Theano
• MinPy a NumPy interface above MXNet backend, in CPU/GPU mode
• Gluon library in Apache MXNet, CPU/GPU mode
• CuPy an open-source matrix library accelerated with NVIDIA CUDA.
• R Project for Statistical Computing

Tasks:

See benchmark.ipynb notebook for details.

Benchmark output on hexacore CPU / Kepler GPU

Task I: Time-series model (GARCH / Scan operation benchmark)

Task II: Monte Carlo simulation (Barrier option pricing)

Task III: Logistic regression using Hybrid Monte Carlo (HMC)