Skip to content

Latest commit

 

History

History
80 lines (53 loc) · 2.74 KB

README.md

File metadata and controls

80 lines (53 loc) · 2.74 KB
Raw

RapidAI

RapidAI is a Python library built on top of PyTorch, designed to streamline the process of training machine learning models. It simplifies workflows by offering modular APIs for model initialization, training, evaluation, and data handling.

Installation

Use the package manager pip to install RapidAI:

pip install rapidai

Make sure you have Python 3.8+ and PyTorch installed before installing RapidAI.

Usage

Below is an example of how to load data, initialize a model, and train it using RapidAI:

from rapidai.core import initialize_weights
from rapidai.datasets import load_mnist
from rapidai.training import train_model

# Load the MNIST dataset
train_loader, val_loader = load_mnist(batch_size=32)

# Define a custom CNN model
model = MyCustomCNN()

# Initialize the model’s weights
initialize_weights(model)

# Train the model for 10 epochs
train_model(model, train_loader, val_loader, epochs=10)

Modules and Functions

Core Utilities (core.py)

  • initialize_weights(model): Initializes the weights for better convergence.
  • train_model(): Manages training loops and backpropagation.

Dataset Handling (datasets.py)

  • load_mnist(batch_size=32): Loads and preprocesses the MNIST dataset.
  • split_data(): Splits data into train, validation, and test sets.

Training Management (training.py)

  • train_one_epoch(): Runs a single epoch of training.
  • evaluate_model(): Evaluates model performance on validation/test data.

Neural Networks (conv.py, resnet.py)

  • conv.py: Functions for building convolutional layers.
  • resnet.py: Pre-built ResNet architectures for transfer learning.

Learner and Optimization (learner.py, accel.py)

  • learner.py: Encapsulates the training process with metric tracking.
  • accel.py: Optimized implementations of SGD for faster convergence.

Data Augmentation and Activation Functions

  • augment.py: Tools for augmenting data with techniques like random flipping.
  • activations.py: Implements ReLU, Sigmoid, Tanh, and other activations.

Example Notebooks

The following Jupyter notebooks in the repository demonstrate the use of RapidAI:

  • 00_core.ipynb: Overview of the core functionalities.
  • 01_minibatch_training.ipynb: Setting up minibatch training workflows.
  • 04_learner.ipynb: Using the learner module for training.
  • 08_resnet.ipynb: Transfer learning with a ResNet model.

Contributing

Pull requests are welcome! For major changes, please open an issue first to discuss what you would like to change.

Make sure to update tests as appropriate when contributing.

License

MIT