smolNN

A lightweight, from-scratch implementation of a Multi-Layer Perceptron (Neural Network) built in Python, designed to classify Palmer Penguins based on physical characteristics. It features a graphical user interface (GUI) for easy tuning of hyperparameters and real-time model evaluation.

Features

• Custom Neural Network from Scratch (nn.py):

  • Implements an MLP without external deep learning frameworks (uses numpy and pandas).
  • Supports multiple optimization algorithms: Stochastic Gradient Descent (SGD), Batch Gradient Descent (BGD), and Adam.
  • Dynamic hidden layers mapping with customizable neuron counts.
  • Activation functions including Sigmoid and Tanh, accompanied by a Softmax output layer for robust multi-class probability extraction.

• Graphical User Interface (interface.py):

  • A friendly desktop application built with tkinter.
  • Modify parameters such as the learning rate, number of epochs, activation function, and hidden layer structure on the fly.
  • Train the model dynamically, evaluate test/train accuracies, and compute confusion matrices.
  • Interactive prediction tab: Input new penguin features (e.g., Culmen Length, Body Mass, Origin Location) and predict the exact species immediately.

• Data Preprocessing Pipeline:

  • Handles missing values safely.
  • Performs One-Hot Encoding for categorical features.
  • Automatic Z-score standardization (normalization) to accelerate convergence.

Project Structure

File	Description
nn.py	The core data structures and Neural Network logic. Includes Grid Search for automatic tuning.
interface.py	The GUI script linking directly to the Neural Network backend (nn.py).
penguins.csv	The Palmer Penguins dataset used to train predictions.
requirements.txt	File containing the necessary Python packages to run the project.
observations.pdf	Documentation with insights and visualizations about data preparation and training methodology.

Getting Started

1. Install Requirements

Make sure you have Python 3.x installed. Then, simply install dependencies through pip:

pip install -r requirements.txt

2. Running the GUI

Launch the comprehensive graphical interface to set up models dynamically:

python interface.py

3. Native Grid Search (CLI)

You can directly execute the neural network file to invoke the built-in grid search and test best parameters:

python nn.py

This prints the optimal configuration matrices and visualizes confusion plots directly.