TESTING_GUIDE.md

🧪 Testing Guide for Multilingual Emotion Classifier

This guide provides comprehensive testing capabilities for the rmtariq/multilingual-emotion-classifier model.

🚀 Quick Start

Installation

# Install requirements
pip install -r requirements_testing.txt

# Or install manually
pip install torch transformers numpy pandas scikit-learn

Basic Usage

# Quick test (recommended for first-time users)
python test_model.py --test-type quick

# Comprehensive test
python test_model.py --test-type comprehensive

# Interactive testing
python test_model.py --test-type interactive

# Performance benchmark
python test_model.py --test-type benchmark

# Run all tests
python test_model.py --test-type all

📋 Test Types

1. 🚀 Quick Test

Purpose: Fast validation of core functionality
Duration: ~30 seconds
Coverage: 13 essential test cases (English + Malay)

python test_model.py --test-type quick

What it tests:

• ✅ Basic English emotions (6 cases)
• ✅ Basic Malay emotions (4 cases)
• ✅ Previously problematic cases (3 cases)

Expected Results: >90% accuracy

2. 🔬 Comprehensive Test

Purpose: Thorough validation across all categories
Duration: ~2 minutes
Coverage: 24 test cases across multiple categories

python test_model.py --test-type comprehensive

Test Categories:

• English Basic: Core English emotion expressions
• Malay Basic: Core Malay emotion expressions
• Malay Fixed Issues: Previously problematic cases (now fixed)
• Edge Cases: Boundary and special cases

Expected Results: >85% overall accuracy

3. 🎮 Interactive Test

Purpose: Manual testing with custom inputs
Duration: User-controlled
Coverage: Unlimited custom test cases

python test_model.py --test-type interactive

Features:

• Real-time emotion classification
• Confidence scoring
• Emoji visualization
• Easy exit (type 'quit')

Example Session:

💬 Your text: I am so excited!
🎭 Result: 😊 happy
📊 Confidence: 99.8%
💪 High confidence!

💬 Your text: Saya gembira!
🎭 Result: 😊 happy
📊 Confidence: 99.9%
💪 High confidence!

4. ⚡ Benchmark Test

Purpose: Performance and speed evaluation
Duration: ~1 minute
Coverage: 100 predictions for timing analysis

python test_model.py --test-type benchmark

Metrics Measured:

• Total processing time
• Average time per prediction
• Predictions per second
• Performance classification

Expected Results: >5 predictions/second

🎯 Supported Emotions

The model classifies text into 6 emotion categories:

Emotion	Emoji	Description	Example (English)	Example (Malay)
anger	😠	Frustration, rage	"I'm so angry!"	"Marah betul!"
fear	😨	Anxiety, worry	"I'm scared!"	"Takut sangat!"
happy	😊	Joy, excitement	"I'm so happy!"	"Gembira sangat!"
love	❤️	Affection, care	"I love you!"	"Sayang kamu!"
sadness	😢	Sorrow, grief	"I'm so sad"	"Sedih betul"
surprise	😲	Amazement, shock	"What a surprise!"	"Terkejut betul!"

🔧 Advanced Usage

Custom Model Testing

# Test a different model
python test_model.py --model "your-model-name" --test-type quick

# Test local model
python test_model.py --model "./path/to/local/model" --test-type comprehensive

Programmatic Usage

from test_model import EmotionModelTester

# Initialize tester
tester = EmotionModelTester("rmtariq/multilingual-emotion-classifier")

# Run specific tests
quick_accuracy = tester.quick_test()
comprehensive_accuracy = tester.comprehensive_test()
speed = tester.benchmark_test()

print(f"Quick test accuracy: {quick_accuracy:.1%}")
print(f"Comprehensive accuracy: {comprehensive_accuracy:.1%}")
print(f"Speed: {speed:.1f} predictions/second")

📊 Expected Performance

Accuracy Targets

• Quick Test: >90% accuracy
• Comprehensive Test: >85% accuracy
• English Performance: >95% accuracy
• Malay Performance: >85% accuracy

Speed Targets

• CPU Performance: >5 predictions/second
• GPU Performance: >20 predictions/second

Confidence Levels

• High Confidence: >90% (💪)
• Good Confidence: 70-90% (👍)
• Low Confidence: <70% (⚠️)

🐛 Troubleshooting

Common Issues

1. Model Loading Errors

❌ Error loading model: ...

Solutions:

• Check internet connection
• Verify model name spelling
• Try: pip install --upgrade transformers

2. CUDA/GPU Issues

CUDA out of memory

Solutions:

• The model automatically falls back to CPU
• Reduce batch size if using custom code
• Use --device cpu flag if available

3. Slow Performance

⚠️ SLOW. Consider optimization.

Solutions:

• Use GPU if available
• Close other applications
• Consider model quantization for production

Getting Help

If you encounter issues:

1. Check Requirements: Ensure all dependencies are installed
2. Update Libraries: pip install --upgrade transformers torch
3. Check Model Status: Visit model page
4. Report Issues: Create an issue on the repository

🎯 Test Case Examples

English Test Cases

# Basic emotions
"I am so happy today!"          # → happy
"This makes me really angry!"   # → anger
"I love you so much!"           # → love
"I'm scared of spiders"         # → fear
"This news makes me sad"        # → sadness
"What a surprise!"              # → surprise

Malay Test Cases

# Basic emotions
"Saya sangat gembira!"          # → happy
"Aku marah dengan keadaan ini"  # → anger
"Aku sayang kamu"               # → love
"Saya takut dengan ini"         # → fear
"Sedih betul dengan berita"     # → sadness
"Terkejut dengan kejadian"      # → surprise

# Fixed issues (previously problematic)
"Ini adalah hari jadi terbaik"  # → happy (was: anger)
"Terbaik!"                      # → happy (was: surprise)
"Ini adalah hari yang baik"     # → happy (was: anger)

📈 Performance History

Version 2.1 (Current)

• ✅ Overall Accuracy: 85.0%
• ✅ English Performance: 100%
• ✅ Malay Performance: 100% (fixed issues)
• ✅ Speed: 5-20 predictions/second

Key Improvements

• 🔧 Fixed Malay birthday context classification
• 🔧 Fixed "baik/terbaik" positive expression recognition
• 🔧 Improved confidence scores
• 🔧 Enhanced robustness

🏆 Success Criteria

A successful test run should show:

• ✅ Quick Test: >90% accuracy
• ✅ No Critical Failures: All basic emotions working
• ✅ Malay Fixes Verified: Birthday/positive contexts → happy
• ✅ Reasonable Speed: >5 predictions/second
• ✅ High Confidence: Most predictions >90%

---

Model Repository: https://huggingface.co/rmtariq/multilingual-emotion-classifier
Author: rmtariq
Last Updated: June 2024