README.md

metadata

title: Emotion Classifier Deploy
emoji: 🚀
colorFrom: red
colorTo: red
sdk: docker
app_port: 8501
tags:
  - streamlit
pinned: false
short_description: Streamlit template space

Multi-Label Emotion Classifier

A fine-tuned RoBERTa-based multi-label emotion classifier with a Streamlit web interface. The model predicts five emotions: anger, fear, joy, sadness, and surprise.

Features

• Multi-label Classification: Detects multiple emotions in a single text
• Fine-tuned RoBERTa-Large: Built on FacebookAI/roberta-large for robust performance
• Interactive Web UI: Streamlit-based interface with adjustable decision threshold
• Batch Inference: Process multiple texts at once
• Visualization: Bar charts showing emotion probabilities

Quick Start

Option 1: Run with Docker

docker build -t emotion-classifier .

docker run -p 8501:8501 emotion-classifier

Open your browser at http://localhost:8501

Option 2: Run Locally

pip install -r requirements.txt

streamlit run src/streamlit_app.py

Model Architecture

• Base Model: FacebookAI/roberta-large
• Classification Head:
  • Linear(1024 → 1024) → ReLU → Dropout(0.2) → Linear(1024 → 5)
• Pooling: Mean pooling with attention mask
• Output: Sigmoid activation for multi-label classification

Usage

Web Interface

1. Enter one or more text lines in the text area
2. Adjust the decision threshold using the sidebar slider
3. Click "Run inference" to get predictions
4. View results in the table and bar chart

Python API

from transformers import AutoModel, AutoTokenizer
from huggingface_hub import hf_hub_download
import torch

# Load model components
repo_id = "JashMevada/emotion-classifier"
tokenizer = AutoTokenizer.from_pretrained(repo_id)
encoder = AutoModel.from_pretrained(repo_id, add_pooling_layer=False)

# Load classifier head
weights_path = hf_hub_download(repo_id, "classifier.pth")
classifier = torch.nn.Sequential(
    torch.nn.Linear(1024, 1024),
    torch.nn.ReLU(),
    torch.nn.Dropout(0.2),
    torch.nn.Linear(1024, 5),
)
classifier.load_state_dict(torch.load(weights_path, map_location="cpu"))

# Inference
text = "I am so happy today!"
encoded = tokenizer(text, return_tensors="pt", truncation=True, max_length=512)

with torch.no_grad():
    outputs = encoder(**encoded).last_hidden_state
    mask = encoded["attention_mask"].unsqueeze(-1).float()
    pooled = (outputs * mask).sum(dim=1) / mask.sum(dim=1).clamp(min=1e-9)
    logits = classifier(pooled)
    probs = torch.sigmoid(logits)

emotions = ["anger", "fear", "joy", "sadness", "surprise"]
for emo, prob in zip(emotions, probs[0]):
    print(f"{emo}: {prob:.3f}")

Links

• Model: HuggingFace Hub
• Demo: Streamlit App