Mini-Keras is a minimalistic reimplementation of the popular Keras deep learning library. This project is designed to provide a deeper understanding of the internal workings of Keras by recreating its basic functionality from scratch.
Mini-Keras includes the following components:
- Layers: Basic building blocks of neural networks, including fully connected (Dense) layers and input layers.
- Activation Functions: Functions that determine the output of a neuron, including Sigmoid, Tanh, ReLU, Leaky ReLU, and Linear.
- Loss Functions: Functions that measure the difference between the predicted and actual values, including Mean Squared Error (MSE) and Categorical Cross-Entropy (CCE).
- Metrics: Functions that evaluate the performance of the model, including MSE and Accuracy.
- Optimizers: Algorithms that update the weights of the model during training, including Stochastic Gradient Descent (SGD).
- Regularizers: Functions that apply penalties on layer parameters during optimization, including L1 and L2 regularization.
An example of how to use Mini-Keras to train a neural network model on the CIFAR-10 dataset is provided in the example.py file. This includes preprocessing the data, defining the model, and fitting the model to the training data with different learning rates.
Please note that Mini-Keras is a simplified version of Keras and does not include all of its features. It is intended for educational purposes only, to provide a deeper understanding of how Keras works under the hood. For serious deep learning projects, we recommend using the full-featured Keras library.