Caispp

About

This package allows for high level ML model creation. It uses Keras with a Tensorflow backend, and was originally created to be used for the curriculum of USC's CAIS++ (Center for AI in Society, Student Branch).

Use Cases

The package currently supports Image Classification.

Installation

To install run pip install caispp. This package uses Tensorflow 2.0.

Example usage

You can see a jupyter notebook with ouputs in the examples/ directory. The notebook runs the code below:

from caispp import ImageDataset, ImageClassifier, Path

path = Path('example_dataset/') # Path to dataset
dataset = ImageDataset(path, show_distribution=True)

classifier = ImageClassifier(dataset)
classifier.train(epochs=10)

classifier.show_history()

classifier.test(show_distribution=True)

Dataset directory structure

├── example_dataset         
│   ├── test
│   │   ├── class1      # Directory with images of class1
│   │   ├── class2      # Directory with images of class2
│   │   └── ...       
│   ├── train
│   │   ├── class1      # Directory with images of class1
│   │   ├── class2      # Directory with images of class2
│   │   └── ...         
│   ├── valid           # Optional validation set    
│   │   ├── class1
│   │   ├── class2
│   │   └── ... 
└──  

Each of the test/, train/, and valid/ directories contain subdirectories for each class. In those subdirectories, put the images files of that class.

Build the package

To build the package run the build.sh script in the directory. The output is stored in dist/.