RohanBh/cs263_project

Profiling differences between Cython and CPython

CS263 Project

Setup

$ git clone https://github.com/RohanBh/cs263_project.git
$ cd cs263_project
$ conda env create -f conda_cs263.yml
$ conda activate conda_cs263

Building and Running the benchmarks

We demonstrate the steps used for running the benchmarks ml-mnist. Other programs require similar steps. In the directory, cs263_project/programs/ml-mnist, we see the following files:

• Profiling.md: Contains information from Profiling the program and comparing performance of different variants and understanding it.
• Run python <setup-script> build_ext --inplace to compile Cython (using setup_cython.py) and Cython Typed (using setup_cython_typed.py) programs and create .so files.
• profile_train.py: Use this program to profile/time the training of a neural network with Numpy.
• Similarly, use profile_train_cython.py and profile_train_cython_typed.py for profiling Cython and Cython-Typed respectively.

The programs profile_*.py run the corresponding benchmark variant for 5 times and prints the execution time and the median time (in seconds).