Multilayer perception implemented from scratch. Standard feedforward neural network to solve classification and regression problems.

• Single/Multioutput Regression
• Binary/Multilabel/Multiclass Classification
• L1/L2 Regularisation
• Dropout Regularisation
• Mini-Batch Gradient Descent (GD)
• Adam Optimisation (Momentum GD + RMS Propagation)
• Learning Rate Decay

The baseline verification was performed with Scikit-Learn's MLPClassifier on a generated dataset (3 features, 4 classes). Test accuracy scores for both models were about 0.9, with comparable runtimes.