# Chess AI

In our midterm course project for the Symbolic AI Course we created an GOFAI system using negamax-search, pruning and iterative deepening and various heuristic. The system is specified to play a smaller variant of chess against other agents. It plays better then an inexperienced chess player.

## Requirements

- Python >= 3.7
- pygame
- tkinter

Install dependencies:

```bash
pip install pygame
pip install tk
```

## Starting the GUI

Run `ChessMain.py` with the following arguments:

```bash
python3 ChessMain.py --agent1 [AGENT] --agent2 [AGENT] --verbose --time_control [SECONDS] --use_gui
```

**Agent options:**

| Value       | Description                                              |
|-------------|----------------------------------------------------------|
| `Human`     | Play yourself via the GUI                                |
| `MrRandom`  | Plays completely random valid moves                      |
| `MrNovice`  | Plays using a basic minimax algorithm                    |
| `Agent1`    | Uses the `Agent` class from `student_agents/template.py` |
| `Agent2`    | Uses the `Agent` class from `student_agents/template2.py`|
| `path/to/agent.py` | Path to a custom agent file (not supported on Windows) |

**Example — Human vs. your agent:**

```bash
python3 ChessMain.py --agent1 Human --agent2 Agent1 --time_control 20 --use_gui
```

**Example — Agent vs. Agent (no GUI):**

```bash
python3 ChessMain.py --agent1 Agent1 --agent2 MrRandom --time_control 20
```

## Board Selection

Edit `Settings.json` and set the board value to `1` or `2` to choose between available board layouts.

## Agent Implementation

Your agent lives in `student_agents/template.py`. The class must be named `Agent` and follow the interface in that file. Use `update_move` to register preliminary moves so your agent does not forfeit on timeout.

Your agent must be **single-threaded**. Multi-threaded agents are disqualified.



## Algorithm

Our agent (`student_agents/template.py`) uses **Negamax with Alpha-Beta pruning** and **iterative deepening**:

- **Negamax**: A symmetric formulation of minimax — each node maximizes the negated score of its children, so white and black are handled with the same code.
- **Alpha-Beta pruning**: Cuts off branches that cannot influence the final decision, dramatically reducing the number of nodes evaluated.
- **Iterative deepening**: The search starts at depth 1 and increments indefinitely. After each completed depth, the best move found so far is registered via `update_move`, so the agent always has a valid move ready when time runs out.

## Evaluation Function

The heuristic scores the board by summing material value and positional bonuses for each piece:

**Material values:** K=0, Q=9, R=5, B=3, N=3, p=1

**Piece-square tables:** Each piece type has a 6×6 positional score table that rewards good squares (e.g. knights in the center, pawns advanced toward promotion). Pawns have separate tables for white and black to incentivise advancement.

**Pawn structure / endgame:** The white and black pawn tables assign very high scores (9–10) to pawns near the promotion rank, encouraging the agent to push pawns in the endgame.

**Zobrist hashing + transposition table:** Board positions are hashed using a Zobrist hash (random 64-bit integers per piece/square combination, XOR'd together). Evaluated positions are cached in a lookup table so repeated positions are not re-evaluated.

---