Skip to content

Instructions

Jump to bottom
Calvin Miracle edited this page Aug 21, 2021 · 2 revisions

How to install and run the FeatureRanking project

Requirements

This project was built and tested on Windows 10 64-bit, with up-to-date 
CPU and GPU drivers. You'll need at least one OpenCL device discoverable on 
your workstation, a "recent" version of Anaconda (eg, 2021.05), a Git 
client like [Git Bash](https://git-scm.com) available at the Anaconda command 
line, and a local directory on your workstation where your code repositories
are housed.

Setup from scratch

- Open the Anaconda CMD prompt into the Conda (base) environment.

- Change to your preferred local repository base directory:
      cd C:\Users\your-user-name\...\your-local-repos>

- Clone the FeatureRanking project GitHub repository:
      git clone https://github.com/cbmira01/FeatureRanking

- Change to the FeatureRanking project root directory:
      cd .\FeatureRanking

- Create the Conda environment for the project:
      conda env create --file environment.yml

- After packages have loaded, activate the new Conda environment:
      conda activate feature_ranking

Test the installation

- To run Python programs, change to the .\src folder:
    cd .\src

- The quick-demo program will dump a list of available OpenCL devices,
  and run a very small OpenCL workload on each one:
      python quick_demo.py

- Some units tests can be run by issuing:
      python run_tests.py

After a successful installation

- Run the main program:
      python feature_ranking.py

Return to previously installed project

- Open the Anaconda CMD prompt into the Conda (base) environment.

- Change to the FeatureRanking project root directory:
      cd C:\Users\your-user-name\...\your-local-repos\FeatureRanking

- Activate the existing Conda environment:
      conda activate feature_ranking

Update environment packages

- To update environment packages after repository changes, change to the
  FeatureRanking project root directory, then:
    git pull
    conda env update --name feature_ranking --file environment.yml

Destroy the environment and project

- To destroy the feature_ranking Conda environment:
    conda deactivate
    conda remove --name feature_ranking --all

- To destroy the FeatureRanking project entirely, first destroy the
  feature_ranking environment, then remove the FeatureRanking folder
  and its contents.

If you have difficulty...

I have tested this project on five workstations of various vintages, and have
always managed to either find or establish an OpenCL device. On a Lenovo
T400 (circa 2008), while I found no GPU device, I did discover that the CPU
was rated as an available OpenCL device. Any computer presenting an OpenCL
device should be able to run this project.

Unfortunately, installing PyOpenCL on Windows is still more an art than a 
science. If you encounter difficulty running the quick test, you may need a 
pip 'wheel' (pre-compiled binary) that supports older OpenCL device drivers. 
Here's how to do it...

- Visit [Pre-compiled OpenCL binaries](http://www.lfd.uci.edu/~gohlke/pythonlibs/#pyopencl)

- Download a wheel appropriate for your OpenCL version, vendor type, and 
  Python version. For example, for PyOpenCL 2021.2.2, Python 3.9, and OpenCL 
  version 1.2, try one of the following wheels:
      pyopencl-2021.2.2+cl12-cp39-cp39-win_amd64.whl
      pyopencl-2021.2.2+cl12-cp39-cp39-win32.whl

- Install the wheel. This action will replace the Conda package version
  of PyOpenCL. Make sure the feature_ranking environment is still
  activated, and that you are positioned in FeatureRanking project root:
      pip install C:\Users\your-user-name\...\Downloads\<wheel-name>.whl

- Try the quick test again. If you still see problems, try other pip wheels 
  as appropriate. It is also possible that you are experiencing driver 
  problems, or that you may have no discoverable OpenCL devices on your 
  workstation.

- I have found the "AMD Accelerated Parallel Processing SDK" (AMD APP SDK)
  can allow an Intel CPU to be recognized as an OpenCL devices. This may be 
  worth investigating if your workstation does not have a GPU. 
  See these links:
      [AMD APP SDK](https://en.wikipedia.org/wiki/AMD_APP_SDK)
      [AMD APP SDK discussion](https://github.com/fireice-uk/xmr-stak/issues/1511)
Clone this wiki locally