Deep learning library from scratch (numpy)

see 1D CNN example in repository

GitHub repo

Supports

  • 1D Convolution (Conv1D)
  • Max & Average Pooling (MaxPool1D, AvgPool1D)
  • Fully Connected Layers (FCNN)
  • Activations: ReLU, LeakyReLU, Sigmoid, Swish
  • Loss Functions: Squared Error Loss (MSE), Binary Cross-Entropy (planned)
  • Forward and Backward Pass: gradients computed by chain rule
  • Flatten Layer: Reshapes feature maps for FCNN
  • Utility Function: Padding, logging