# ================================================ # sentiment-analyzer (Python 3.12) # ================================================ # Goal: Classify free‑form customer feedback as Positive / Negative / Neutral with confidence. # Approach: Use Hugging Face transformers pipeline with a domain‑tuned RoBERTa model. # ================================================ # Folder layout (all files defined below): # ├─ requirements.txt # ├─ README.md # └─ src # ├─ config.py ← runtime settings # ├─ model.py ← wraps the HF pipeline # ├─ cli.py ← command‑line batch analyzer # ├─ service.py ← optional FastAPI microservice # └─ utils.py ← helper for CSV/JSON I/O # ================================================ # ================================================ # requirements.txt # ================================================ transformers==4.40.1 torch>=2.1.2 # CPU‑only wheel will auto‑install pandas==2.2.2 fastapi==0.110.2 uvicorn==0.29.0 scikit-learn==1.4.2 # not strictly needed but handy for metrics python-dotenv==1.0.1 rich==13.7.1 # ================================================ # README.md (excerpt) # ================================================ # 1. Install & run CLI # python -m venv .venv && source .venv/bin/activate # pip install -r requirements.txt # python -m src.cli --input reviews.csv --text-column comment # # 2. Start REST API # uvicorn src.service:app --reload --port 8000 # # 3. Example cURL # curl -X POST http://localhost:8000/analyze -d '{"text":"Loved the product!"}' -H "Content-Type: application/json" # ================================================ # src/config.py # ================================================ from pathlib import Path from functools import lru_cache import os from dotenv import load_dotenv BASE_DIR = Path(__file__).resolve().parent load_dotenv(BASE_DIR.parent / '.env', override=True) MODEL_NAME = os.getenv( 'HF_MODEL', 'cardiffnlp/twitter-roberta-base-sentiment' # 3‑class POS/NEU/NEG ) DEVICE = 0 if os.getenv('GPU', 'false').lower() == 'true' else -1 # -1 ⇒ CPU BATCH_SIZE = int(os.getenv('BATCH_SIZE', 32)) # ================================================ # src/model.py # ================================================ """Centralised model loader so the weights are loaded once per process.""" from transformers import AutoTokenizer, AutoModelForSequenceClassification, pipeline from config import MODEL_NAME, DEVICE @lru_cache def get_pipeline(): tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME) model = AutoModelForSequenceClassification.from_pretrained(MODEL_NAME) return pipeline('sentiment-analysis', model=model, tokenizer=tokenizer, device=DEVICE) def predict(texts: list[str]): """Return list of dicts: {label: str, score: float}.""" pipe = get_pipeline() return pipe(texts, truncation=True, batch_size=len(texts)) # ================================================ # src/utils.py # ================================================ import pandas as pd from rich.console import Console from rich.table import Table console = Console() def load_texts(path: str | os.PathLike, column: str = 'text') -> list[str]: df = pd.read_csv(path) if column not in df.columns: raise ValueError(f"Column '{column}' not found in CSV") return df[column].astype(str).tolist(), df def render_preview(preds: list[dict], texts: list[str], k: int = 10): table = Table("Text", "Label", "Confidence", title="Sample Sentiment Preview") for t, p in zip(texts[:k], preds[:k]): table.add_row(t[:60] + ('…' if len(t) > 60 else ''), p['label'], f"{p['score']:.2f}") console.print(table) # ================================================ # src/cli.py # ================================================ """Run sentiment analysis on a CSV and append predictions.""" import argparse, json from model import predict from utils import load_texts, render_preview import pandas as pd parser = argparse.ArgumentParser(description="Batch sentiment analyzer") parser.add_argument('--input', required=True, help='CSV file with text column') parser.add_argument('--text-column', default='text', help='Column name containing review text') parser.add_argument('--output', default='output.csv', help='Destination CSV (default: output.csv)') args = parser.parse_args() texts, df = load_texts(args.input, args.text_column) preds = predict(texts) labels = [p['label'] for p in preds] scores = [p['score'] for p in preds] df['sentiment'] = labels df['confidence'] = scores df.to_csv(args.output, index=False) render_preview(preds, texts) print(f"✅ Analysis complete – results saved to {args.output}") # ================================================ # src/service.py # ================================================ from fastapi import FastAPI, HTTPException from pydantic import BaseModel from model import predict app = FastAPI(title="Sentiment Analyzer API") class TextIn(BaseModel): text: str class PredictionOut(BaseModel): label: str score: float @app.post('/analyze', response_model=PredictionOut) async def analyze(payload: TextIn): try: res = predict([payload.text])[0] return {'label': res['label'], 'score': res['score']} except Exception as e: raise HTTPException(status_code=500, detail=str(e)) --------------------------------------------------------------------------------------------------------------------------- A turnkey **Python “sentiment-analyzer” codebase**—with CLI and optional REST API—is now in the canvas. It uses a RoBERTa model fine-tuned on 3-class sentiment data, giving **Positive / Neutral / Negative** plus confidence scores. ### External stack | Dependency | Purpose | |------------|---------| | **transformers + torch** | Load/run the pretrained model | | **pandas** | Batch CSV ingestion & export | | **rich** | Pretty CLI table preview | | **fastapi + uvicorn** | Lightweight HTTP endpoint | | **python-dotenv** | Override model/device settings via `.env` | | **scikit-learn** | (Optional) evaluation utilities | All pinned in **requirements.txt**. ### Quick-start (CLI) ```bash git clone <repo> cd sentiment-analyzer python -m venv .venv && source .venv/bin/activate pip install -r requirements.txt # run on your reviews CSV (needs a “comment” column) python -m src.cli --input reviews.csv --text-column comment --output scored.csv ``` You’ll get a rich preview in-terminal and a new CSV with **sentiment** and **confidence** columns. ### Run as an API ```bash uvicorn src.service:app --reload --port 8000 ``` POST JSON like: ```json { "text": "The support team solved my issue in minutes—fantastic!" } ``` and receive: ```json { "label": "POS", "score": 0.97 } ``` ### Config tweaks * **Model** Set `HF_MODEL` in `.env` to any Hugging Face checkpoint supporting 3-class sentiment (e.g. `nlptown/bert-base-multilingual-uncased-sentiment` for 5-star grading). * **GPU** If CUDA is available, set `GPU=true` to run on device 0. * **Batch size** Override `BATCH_SIZE` in `.env` for memory tuning. ### Deployment ideas | Target | Steps | |--------|-------| | **Docker** | `FROM python:3.12-slim`, copy files, `pip install`, expose 8000, `CMD uvicorn src.service:app --host 0.0.0.0 --port 8000` | | **AWS Lambda** | Package with `serverless` + `transformers[torch]`, enable container image if size > 250 MB. | | **Streamlit** | Swap FastAPI for Streamlit to build an interactive demo UI in minutes. | Open the canvas to inspect, customize, or extend—happy to add batch metrics, HTML reporting, or multilingual support if that would help!