# the-fashion-queen-agent

**Repository Path**: suanfamama/the-fashion-queen-agent

## Basic Information

- **Project Name**: the-fashion-queen-agent
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 1
- **Forks**: 0
- **Created**: 2025-02-04
- **Last Updated**: 2025-02-10

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# an AI Agent with Wisdom Graph for capture the beauty in fashion

## Intro
* 我们研发了一个在时尚领域能捕捉美的人工智能智能体
* 请在小红书平台上搜索"原优舍买手店"以给程序员群体选择适合且时尚的服饰穿搭。
* 我们团队目前还处于挺初级的状态，欢迎大佬提出改进建议。
* 相关论文：https://github.com/algmon/the-fashion-queen-agent/blob/main/agent.pruning.202501.pdf
* ICML 2025 OpenReview Profile: https://openreview.net/forum?id=akfdCT2lWP

![](./cover.photo.png)
* AI智能体帮助我们捕捉美，整理美及在小红书平台上展示美

## How to run the demos?

1. Install dependencies:
```bash
pip install -r requirements.txt
```

2. Run demo 1:
```bash
python play.demo.1.py
```

3. Run demo 2:
```bash
python play.demo.2.py
```

## How to Test?

1. Run all tests:
```bash
python test.py
```

2. Run specific test categories:
```bash
# Test environment
python test.py env init
python test.py env all

# Test agent components
python test.py agent init
python test.py agent perception
python test.py agent planning
python test.py agent reasoning
python test.py agent action
```

Test categories:
- `env`: Environment tests
- `agent`: Agent component tests
  - `init`: Initialization tests
  - `action`: Action space tests
  - `planning`: Planning capabilities
  - `perception`: Perception system
  - `reasoning`: Decision making

## the Proposed Conceptual Model
### the cognitive cycle for an AI agent
![](./cycle.svg)

## the Proposed Mathematical Model
### 1. Model the Perception Space (See paper Section 3.1 for detail)
- Environment analysis
- State estimation
- Noise handling
- Sensor fusion

### 2. Model the Planning Space (See paper Section 3.2 for detail)
- Implemented Hierarchical Task Networks (HTNs) for task decomposition
- Cost function: C(t) = Σ w_i * f_i(t) considering:
  - Resource requirements R(t)
  - Estimated time T(t)
  - Failure risk F(t)
  - Task importance weights w_i

Example task decomposition:
```python
high_level_task = TaskNode(
    task_id="navigate_to_target",
    task_type="compound",
    description="Navigate to target position"
)
```

### 3. Model the Reasoning Space (See paper Section 3.3 for detail)
- Bayesian Network Implementation
- Probabilistic decision making
- Resource consideration
- Environmental condition assessment
- Uncertainty handling
- CPD optimization

### 4. Model the Action Space (See paper Section 3.4 for detail)
- Q-Learning based navigation
- State-action value learning
- Epsilon-greedy exploration
- Reward shaping for 3D navigation
- Experience-based decision making

## the Proposed Pruning Algorithm
-  Wisdom Graph based
![](./without.mama.pruning.png)
- without Mama Pruning
![](./with.mama.pruning.png)
- with Mama Pruning

## the Proposed AI Agent Architecture
- brain
- body

## Prelimary Implementation and Experimental Results
![](./alg.qlearning.viz.revisions/alg.qlearning.draft.4.png)
- a basic ai agent for path planning via the perception, planning, reasoning and action cycle
- demo 1

![](./data/env.4.jpg)
![](./results/demo.2.step.1.png)
![](./results/demo.2.step.2.png)
![](./results/demo.2.step.3.png)
![](./results/demo.2.step.4.png)
- an ai agent capture the beauty from visual signals via fast cognitive cycle
- demo 2

![](./data/audio.1.wav)
![](./results/demo.3.step.1.png)
![](./results/demo.3.step.2.png)
![](./results/demo.3.step.3.png)
![](./results/demo.3.step.4.png)
- an ai agent capture the beauty from audio signals via fast cognitive cycle (Demo 3)

**Note:** Demo 3 establishes a framework for processing audio signals. Audio features—such as MFCCs, chroma, spectral contrast, and tonnetz—are extracted using `librosa` from the normalized waveform. The current results (with classical music input) show that while the system successfully extracts these features, the overall beauty evaluation score is low. Further tuning and research on the audio aesthetic metrics are needed to effectively capture "beauty" from audio signals.

## Key Features

### 1. Physical Constraints
- Gravity effects
- Energy costs for different movements
- Maximum jump/climb height
- Ground support requirements
- Terrain and platform interactions

### 2. Navigation Capabilities
- Full 3D movement
- Energy-efficient pathfinding
- Obstacle avoidance
- Dynamic path replanning

### 3. Resource Management
- Energy tracking
- Stamina system
- Recharge planning
- Cost-benefit analysis

### 4. Learning and Adaptation
- Experience accumulation
- Q-value updates
- Dynamic exploration rates
- Performance optimization

## Test Suite
Comprehensive testing framework including:
- Action space testing
- Planning space validation
- Reasoning capability assessment
- Physical constraint verification

## Visualization
- 2D path visualization
- 3D trajectory plotting
- Learning metrics display
- Real-time performance monitoring

## Future Improvements
1. Enhanced terrain interaction
2. More sophisticated path planning
3. Advanced resource optimization
4. Improved learning algorithms

## Demo

## Research Metrics
- Planning efficiency
- Resource utilization
- Learning convergence
- Task completion rates

## Dependencies
- NumPy
- Matplotlib
- NetworkX
- Pandas
- pgmpy (for Bayesian networks)

## Project Structure
```
project/
├── agent.py      # Core agent implementation
├── env.py        # Environment simulation
├── config.py     # Configuration settings
├── spaces/       # Cognitive spaces
│   ├── perception.py
│   ├── planning.py
│   ├── reasoning.py
│   └── action.py
└── test.py       # Testing framework
```

## Next Steps
1. Enhance navigation capabilities
2. Improve resource management
3. Add more environmental features
4. Expand learning capabilities

## One more thing - Happy Chinese New Year of Snake
![](./demo.game.snake.UI.png)