Qnumber

Qnumber

Qnumber is a desktop application for creating, managing, and training handwritten digit datasets with a fully interactive visual workflow.

It is designed as both:

• an educational tool for understanding how neural networks work internally
• a practical desktop environment for preparing 32×32 digit datasets, training a custom model, and testing predictions

The project combines a custom PyQt6 interface with a lightweight NumPy-based neural network engine. The goal is not to hide the mechanics behind abstraction, but to make the full pipeline visible: drawing, dataset preparation, training, inference, and inspection.

---

Features

Dataset creation

• Draw 32×32 digit images manually
• Create training and test samples
• Support for multiple visual formats and controlled experiments
• Fast sample generation workflow
• Reindex dataset files consistently

Dataset management

• Built-in dataset gallery
• Train and test split organization
• Sorting and browsing tools
• Delete-all actions with confirmation
• Clear file naming workflow for large datasets

Training

• Custom feed-forward neural network engine built with NumPy
• Configurable dense architectures such as:
  • 1024 → 128 → 64 → 10
  • 1024 → 64 → 10
• Real training workflow instead of black-box wrappers
• Educational visibility into the pipeline

Testing and inference

• Run predictions on custom samples
• Inspect model output behavior
• Useful for understanding classification confidence and failure cases

Interface

• Desktop UI built with PyQt6
• Modular page-based structure
• Designed for a dark, high-contrast workflow
• Built for direct experimentation rather than notebook-only usage

---

Project Philosophy

Most beginner ML tools abstract away too much. Qnumber does the opposite.

This project is built to let the user see the actual structure of a small neural network system:

1. create or collect image data
2. organize the dataset
3. train a model
4. test the model
5. inspect what happens

That makes Qnumber useful not only as a utility, but as a learning instrument for machine learning fundamentals.

---

Installation

Clone the repository:

git clone https://github.com/milord-x/Qnumber.git
cd Qnumber

Create and activate a virtual environment.

Bash or Zsh:

python3 -m venv .venv
source .venv/bin/activate

Fish:

python3 -m venv .venv
source .venv/bin/activate.fish

Install dependencies:

pip install -r requirements.txt

Run the application:

python3 app.py

---

Requirements

№	library	Version
1	Python	>=3.11+
2	PyQt6	>=6.6
3	numpy	>=1.26
4	Pillow	>=10.0
5	matplotlib	>=3.8

---

Project Structure

Qnumber/
├── app.py
├── requirements.txt
├── assets/
│   └── archive/
├── dataset/
│   ├── train/
│   └── test/
├── ui/
│   ├── main_window.py
│   ├── pages/
│   └── widgets/
├── LICENSE
└── README.md