https://github.com/imtiazsalaf-01/svaraai-reply-classification-pipeline
A complete ML pipeline for classifying email replies into positive, negative, or neutral. Features baseline (Logistic Regression) and transformer (DistilBERT) models, evaluation with Accuracy & F1, and deployment via FastAPI /predict API with Docker support. Built for SvaraAI Internship.
https://github.com/imtiazsalaf-01/svaraai-reply-classification-pipeline
distilbert docker email-reply-classifier fastapi internship-project logistic-regression machine-learning nlp reply-classification svaraai svaraai-reply-classification text-classification
Last synced: 11 days ago
JSON representation
A complete ML pipeline for classifying email replies into positive, negative, or neutral. Features baseline (Logistic Regression) and transformer (DistilBERT) models, evaluation with Accuracy & F1, and deployment via FastAPI /predict API with Docker support. Built for SvaraAI Internship.
- Host: GitHub
- URL: https://github.com/imtiazsalaf-01/svaraai-reply-classification-pipeline
- Owner: Imtiazsalaf-01
- Created: 2025-09-22T02:15:09.000Z (14 days ago)
- Default Branch: main
- Last Pushed: 2025-09-22T02:49:14.000Z (14 days ago)
- Last Synced: 2025-09-22T04:25:43.387Z (14 days ago)
- Topics: distilbert, docker, email-reply-classifier, fastapi, internship-project, logistic-regression, machine-learning, nlp, reply-classification, svaraai, svaraai-reply-classification, text-classification
- Language: Python
- Homepage:
- Size: 792 KB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# SvaraAI-Reply-Classification-Pipeline
End-to-end ML pipeline for classifying email replies as positive, negative, or neutral. Implements baseline (Logistic Regression) and transformer (DistilBERT) models, evaluated with Accuracy & F1, and deployed via FastAPI /predict API with Docker support.
## 🎯 Project Overview
This project implements a complete ML pipeline for email reply classification:
- **Part A**: ML/NLP Pipeline with baseline and transformer models
- **Part B**: FastAPI deployment with `/predict` endpoint
- **Part C**: Analysis and reasoning (see `answers.md`)
## 📊 Dataset
The dataset consists of labeled email replies, categorized into:
- **Positive**: Interested, excited, wants to proceed
- **Negative**: Not interested, wants to be removed
- **Neutral**: Asking questions, requesting information
## 🚀 Quick Start
### Prerequisites
- Python 3.8+
- pip
### Installation & Setup
1. **Install dependencies**
```bash
pip install -r requirements.txt
```
2. **Run the complete ML pipeline**
```bash
python notebook.py
```
This will:
- Load and preprocess the dataset
- Train baseline model (TF-IDF + Logistic Regression)
- Optionally train transformer model (DistilBERT)
- Evaluate and compare models
- Save trained models
3. **Start the API server**
```bash
python src/app.py
```
The API will be available at `http://localhost:8000`
4. **Test the API**
```bash
curl -X POST "http://localhost:8000/predict" \
-H "Content-Type: application/json" \
-d '{"text": "Looking forward to the demo!"}'
```
Expected response:
```json
{
"label": "positive",
"confidence": 0.87
}
```
## 📁 Project Structure
```
├── notebook.py # Main ML pipeline (Part A)
├── src/
│ └── app.py # FastAPI application (Part B)
├── data/
│ ├── reply_classification_dataset.csv
│ ├── train.csv
│ ├── val.csv
│ └── test.csv
├── models/
│ ├── baseline_model.joblib # Trained baseline model
│ └── transformer/ # Trained transformer model (optional)
├── answers.md # Short answers (Part C)
├── requirements.txt
├── Dockerfile
└── README.md
```
## 🔧 API Usage
### Endpoint: POST /predict
**Input:**
```json
{
"text": "Looking forward to the demo!"
}
```
**Output:**
```json
{
"label": "positive",
"confidence": 0.87
}
```
### Python Example
```python
import requests
response = requests.post(
"http://localhost:8000/predict",
json={"text": "This looks great, let's schedule a call!"}
)
print(response.json())
# Output: {"label": "positive", "confidence": 0.92}
```
## 📈 Model Performance
### Baseline Model (TF-IDF + Logistic Regression)
- **Accuracy**: ~99%
- **F1 Score**: ~99%
- **Inference Time**: ~1-5ms per sample
- **Model Size**: ~34KB
### TTransformer Model (DistilBERT) – Advanced
- **Accuracy**: Typically 1-3% higher than baseline
- **Inference Time**: ~20-50ms per sample
- **Model Size**: ~250MB
## 🐳 Docker Deployment
```bash
# Build the image
docker build -t reply-classifier .
# Run the container
docker run -p 8000:8000 reply-classifier
```
## 🛠️ Development
### Running Individual Components
```bash
# Run only the ML pipeline
python notebook.py
# Run only the API
python src/app.py
# Check API documentation
# Visit http://localhost:8000/docs
```
### Model Selection
The API automatically uses the best available model:
1. Transformer model (if available in `models/transformer/`)
2. Baseline model (fallback)
## 📋 Assignment Deliverables
✅ **Part A - ML Pipeline**: `notebook.py`
- Data preprocessing and exploration
- Baseline model training (TF-IDF + Logistic Regression)
- Transformer model training (DistilBERT)
- Model evaluation and comparison
✅ **Part B - API Deployment**: `src/app.py`
- FastAPI service with `/predict` endpoint
- JSON input/output as specified
- Dockerfile for containerization
✅ **Part C - Analysis**: `answers.md`
- Short answer questions about model improvement, bias prevention, and prompt design
## 🚨 Troubleshooting
**Model not found error:**
```bash
# Make sure to run the ML pipeline first
python notebook.py
```
**API not starting:**
```bash
# Check if port 8000 is available
python src/app.py
```
**Transformer training issues:**
- Comment out transformer training in `notebook.py` if you have limited resources
- The baseline model alone achieves excellent performance
## 📚 Key Features
- **High Performance**: 99% accuracy on test set
- **Fast Inference**: Sub-5ms predictions with baseline model
- **Production Ready**: FastAPI with automatic documentation
- **Containerized**: Docker support for easy deployment
- **Extensible**: Easy to add new models or features
---
**Assignment Alignment**: Fully implements all deliverables (Part A, Part B, Part C).
**Estimated Runtime**: 5-10 minutes for full pipeline
**API Documentation**: Available at `http://localhost:8000/docs` when running