autolens

Purpose

Effortlessly implementation of pre-existing code from open-source Augmented ML and Automated ML tools that support image classification tasks

Usage

This project can be both used as a library and CLI tool.

Dataset preparation

• Main folder with sub-folders, each named with an integer number

Library Setup

• clone repo: git clone https://github.com/AnaMiguelRodrigues1/autolens.git
• move to the root of the project
• install the library: python3.9 setup.py install
• install python virtual environment on the root folder: python -m venv {automl_tool}_venv
• run source -m {automl_tool}_venv/bin/activate

from autolens.LUDWIG.run import main #prior selection of automl tool

main(
    "../../chest_xray/", #dataset path 
    1, #bigger steps for less computational resources
    (255, 255), #target size
    0.2, #size of testing dataset
    0.1 #size of validation dataset
)

CLI Interaction

• clone repo: git clone https://github.com/AnaMiguelRodrigues1/autolens.git
• move to the root of the project

python3.9 autolens.py "ludwig" "../../chest_xray"
  --target_size "(255, 255)"
  --test_percentage "0.2"
  --val_percentage "0.1"
  --clean_metadata "store_true"
  --cache_dir "{home_dir}/.cache/autolens"

Configuration Details

S.F. - Supported Framework I.S. - Interface Solutions Lang. - Programming Language O.S. - Operative System

	Fastai v2.7.12	Ktrain v0.37.2	Ludwig v0.8.1.post1	Autogluon1 v0.8.2	Autokeras v1.1.0
S.F.	Pytorch v1.13.1	Tensorflow v2.11	Tensorflow2	Pytorch v1.13.1	Tensorflow v>=2.8.03
I.S.	API	API	API/CLI	API	API
Lang.	Python v3.7-v3.10	Python v3.6-v3.10	Python v>=3.8	Python v3.8-v3.10	Python v3.8-v3.11
O.S.	Linux, Windows	Linux	Linux, Windows	Linux, Windows4	Linux, Windows5, MacOS

More Information

• AutoKeras: ImageClassifier
• AutoGluon: AutoMM
• Ludwig: LudwigModel
• Ktrain: vision
• Fastai: vision_learner

Footnotes

1. Uses Fastai as one of the installation requirements. ↩

2. Uses Tensorboard v2.14, a visualization toolkit from Tensorflow. ↩

3. Tensorflow v2.9.1 most compatible with the remaining dependencies. ↩

4. Advisable to use Anaconda. ↩

5. Requires Microsoft Visual C++ and v>7. ↩