search

WayScience / CytoSnake-Benchmarks

CytoSnake benchmarks
1 stars 1 forks source link

readme

CytoSnake-Benchmarks

Welcome to the CytoSnake's Benchmark Repository, where we conduct version-controlled performance tracking of CytoSnake's workflows. This repository benchmarks various image-based profiling processing workflows. We use this repo as a version control system to maintain transparency and track changes in performance. Below we document our benchmarking process, including the datasets used, configuration files, performance metrics measured, and the selected workflow benchmarked Python notebooks.

Structure

image

CytoSnake-Benchmark repository structure

The image above provides an overview of the CytoSnake-Benchmark repository structure. The central elements include the src/ directory (red), housing utility functions for processing raw benchmark files and allowing those functions to be modular. Adjacent is the all-benchmarks/ directory (green), acting as a central repository for benchmarks conducted on specific datasets (blue). Within each tested benchmark folder (yellow) (e.g inside benchmark name 1) a jupyter notebook is used to process the benchmark files and generates plots in the image/ dir and benchmarking results in the results/ dir. Each folder features a detailed README specifying test contents and datasets used.

Below is a table that describes all of the currently available benchmarks inside the all-benchmarks directory: Directory Name Description
cell-health-cp-cp_process Benchmarks the cp_process workflow using the cell-health dataset cell profile features

Installation and Usage

Installation

To get started, follow these steps:

  1. Clone the benchmark repository to your local machine:

    git clone git@github.com:WayScience/CytoSnake-Benchmarks.git

  2. To set up the required dependencies, create a conda environment by utilizing the cytosnake_benchmarks.yaml file.

    conda env create -f env.yaml

    Note: If you already have mamba installed, you can substitute conda with mamba. It is advised to use mamba as it is widely recognized for being a faster alternative to conda. If you haven't installed mamba yet, you can follow the installation instructions here.

  3. Then install the CytoSnake-Benchmark into the local Python environment:

    pip install -e .

    This will give the notebooks to have access to all the functions within the src/ directory.

Creating a benchmark

To create a benchmark, follow these steps: First, create a folder in the all-benchmarks/ directory with the structure {data_type}_{features}_benchmarks/ for the file name. Next, transfer all the necessary files into this folder, which should include the benchmark generated when executing CytoSnake's benchmarking mode. Utilize notebooks to document and display all the raw code and figures generated during the benchmarking analysis. Lastly, it's recommended to store all intermediate files and figures in the repository, allowing others to review and access them.