Skip to content

Latest commit

 

History

History
22 lines (12 loc) · 867 Bytes

README.md

File metadata and controls

22 lines (12 loc) · 867 Bytes
Raw

Neural Network (MLP)

Neural network with fully connected layers, built from scratch.

  • Optimizers: Stochastic Gradient Descent (SGD), L-BFGS

  • Minibatch learning (as generalization for both online learning and batch learning), Momentum and moving average

  • Loss function: mean squared error (MSE)

  • Activation functions: linear, sigmoid, softplus

  • Regularization: L2 (Tichonov), early stopping

  • Metrics: accuracy, mean euclidean error (MEE)

  • Validation techniques: grid search, k-fold cross validation (with parallelization)

  • Utilities: rescaling, 1-of-K encoding, plot routines, read/write routines for dataset where the project has been tested on

This project has been made as part of Machine Learning and Computational Mathematics courses at the University of Pisa. MainCup.py refers to a private contest whose dataset has not been disclosed.