# Design Pattern Detection Model

This model detects software design patterns in Java source code using CodeBERT. The model has been fine-tuned for single-label classification tasks and supports the following design pattern labels:

## Supported Labels

| Label ID | Design Pattern     |
|----------|--------------------|
| 0        | Observer           |
| 1        | Decorator          |
| 2        | Adapter            |
| 3        | Proxy              |
| 4        | Singleton          |
| 5        | Facade             |
| 6        | AbstractFactory    |
| 7        | Memento            |
| 8        | FactoryMethod      |
| 9        | Prototype          |
| 10       | Visitor            |
| 11       | Builder            |
| 12       | Unknown            |

## How to Use

```python
from transformers import AutoTokenizer, AutoModelForSequenceClassification

# Load the model and tokenizer
tokenizer = AutoTokenizer.from_pretrained("ichsanbudiman/design-pattern-detection-codebert")
model = AutoModelForSequenceClassification.from_pretrained("ichsanbudiman/design-pattern-detection-codebert")

# Example input
input_code = """
public class Singleton {
    private static Singleton instance;

    private Singleton() {}

    public static Singleton getInstance() {
        if (instance == null) {
            instance = new Singleton();
        }
        return instance;
    }
}
"""

# Tokenize the input
inputs = tokenizer(input_code, return_tensors="pt", padding="max_length", truncation=True, max_length=512)

# Make predictions
with torch.no_grad():
    outputs = model(**inputs)

# Get the predicted class and label
predicted_class = torch.argmax(outputs.logits, dim=1).item()
predicted_label = model.config.id2label[predicted_class]

print(f"Predicted label: {predicted_label}")
```

## Input Requirements
- **Input Format**: Java code snippets as strings.
- **Max Length**: Input code longer than 512 tokens will be truncated.
- **Padding**: Automatically pads to 512 tokens for batch processing.

## Task
This model performs single-label classification for the detection of design patterns in Java source code. The supported design patterns are listed above.

## Fine-Tuning Details
- **Base Model**: [CodeBERT](https://huggingface.co/microsoft/codebert-base)
- **Dataset**: Fine-tuned on a curated dataset of labeled Java code examples. The dataset was sourced from the following research article:

  > Najam Nazar, Aldeida Aleti, Yaokun Zheng, Feature-based software design pattern detection, Journal of Systems and Software, Volume 185, 2022, 111179, ISSN 0164-1212, [https://doi.org/10.1016/j.jss.2021.111179](https://doi.org/10.1016/j.jss.2021.111179).

- **Metrics**: The model achieves high accuracy on detecting design patterns, making it suitable for software engineering tasks.

## Contact
For inquiries or feedback, please reach out to [Ichsan Budiman](mailto:budimanichsan@gmail.com).

## License
This model is licensed under the Apache 2.0 License.