Multi-Layer Perceptron (MLP)

This project implements and evaluates multilayer perceptrons (MLPs) from scratch using Python and NumPy. It includes code for training on toy datasets such as XOR, as well as comparison with a log-linear model.

Setup

Prerequisites

• Python 3.x
• NumPy

You can install the dependencies using:

pip install -r requirements.txt

Running the code

Training a Model

To train the MLP on provided data:

python train_mlp1.py

To compare with the log-linear model:

python train_loglin.py

Testing on XOR

To test MLP learning on XOR:

python xor_data.py
python train_mlp1.py --data xor

Running Gradient Check

Ensure your gradients are implemented correctly:

python grad_check.py

Evaluation

Model predictions are saved in test.pred. You can evaluate using provided scripts or compare with ground truth labels in the test folder.

Notes

• All models are implemented from scratch using NumPy, with no deep learning frameworks used.