[ADD]: protype with Classifier, encoder, model, problem_embedding, tr…

[divisio74]

…ansformers + testscripte

Description

This PR introduces Prototype v1 of the GNN + Transformer architecture for solving graph optimization problems (MaxCut, Vertex Cover, Independent Set, Graph Coloring).

Architecture Overview

Graph → GNN Encoder → E_local, E_global
Problem ID → Lookup Table → E_prob
Concat [E_global || E_local || E_prob] → Transformer → Classifier → Cross-Entropy

Components

File	Role
encoder.py	GNN (GAT) that produces node embeddings and graph embedding
problem_embedding.py	Learnable lookup table that converts problem ID to vector
transformer.py	Concatenates embeddings + self-attention between nodes
classifier.py	Multi-class classifier with Cross-Entropy loss
model.py	Complete QuantumGraphModel assembling all components
test_model.py	Full test suite (8/8 tests passing)

Key Features

• Unified architecture for multiple problem types (binary and multi-class)
• Cross-Entropy loss enables both 2-class (MaxCut) and k-class (Graph Coloring) problems
• Backpropagation updates all components including the problem embedding lookup table
• Variable graph sizes supported (tested with 4, 8, 16, 32 nodes)
• 3.3M parameters total

Changes include

• Bugfix (non-breaking change that solves an issue)
• New feature (non-breaking change that adds functionality)
• Breaking change (change that is not backwards-compatible and/or changes current functionality)

Checklist

• I have assigned this PR to myself
• I have added at least 1 reviewer
• I have added the needed labels
• I have linked this PR to an issue
• I have linked this PR to a milestone
• I have linked this PR to a project
• I have tested this code
• I have added / updated tests (unit / functional / end-to-end / ...)
• I have updated the README and other relevant documents (guides...)
• I have added sufficient documentation both in code, as well as in the READMEs

Additional comments

This is a clean prototype implementation based on team research discussions. Next steps:

• DataLoader for training data
• Training loop implementation
• Evaluation metrics