Xircuits Comprehensive Keras Library

This comprehensive library for Xircuits encompasses a wide range of functionalities involving Keras datasets, model management, model instantiation, prediction, and transfer learning. It's designed to facilitate various machine learning tasks within the Xircuits visual programming interface.

Prerequisites

• Python 3.8 or higher
• TensorFlow 2.x

Installation

xircuits install tensorflow_keras

You may also install it manually via

pip install -r requirements.txt

Keras Dataset and Model Management

For tasks from dataset preparation to model training and evaluation.

Components

• ReadKerasDataSet: Loads Keras datasets or creates datasets from directories. Supports MNIST, CIFAR-10/100, and custom datasets.
• FlattenImageData: Converts 2D dataset tuples to 1D, suitable for 1D neural networks.
• TrainTestSplit: Splits datasets into training and testing sets with configurable parameters.
• KerasCreate1DInputModel: Creates 1D Keras models for 1D input datasets.
• KerasCreate2DInputModel: Assembles 2D Keras models, perfect for image-based datasets.
• KerasTrainImageClassifier: Trains Keras models for image classification.
• KerasEvaluateAccuracy: Evaluates Keras models against datasets for accuracy and loss.
• ShouldStop: Decides if training should stop based on accuracy targets or max retries.
• SaveKerasModel: Saves Keras models as .h5 files for later use.

Keras Model Instantiation and Prediction

These components support a range of model architectures for image classification.

Components

• LoadKerasModel: Loads Keras models with customizable configurations.
• KerasPredict: Performs predictions with Keras models on images.
• ResNet50, ResNet101, ResNet152: Instantiates various ResNet models with customizable configurations.
• VGG16, VGG19: Provides VGG model architectures for image classification.
• Xception: Implements the Xception architecture for image classification.
• MobileNet: Offers MobileNet architecture with adjustable parameters.

TensorFlow Keras Transfer Learning

Designed for easy integration into Xircuits workflows for transfer learning scenarios, suitable for both beginners and experienced ML practitioners.

Components

• KerasTransferLearningModel: Fetches TensorFlow Keras Models for transfer learning.
• TFDataset: Retrieves datasets from TensorFlow Datasets.
• TrainKerasModel: Trains compiled Keras models with training data.
• TFDSEvaluateAccuracy: Evaluates Keras models' accuracy using TensorFlow Datasets.
• KerasModelCompiler: Compiles TensorFlow Keras models with custom configurations.
• SaveKerasModel: Saves TensorFlow Keras models for future use.