ModelScript

An interactive domain-specific language (DSL) for defining, training, and evaluating machine learning models with real neural network capabilities.

🚀 What's New - Interactive ML Environment

ModelScript now features a complete interactive development environment with:

• 🔥 Interactive REPL: Build and test models in real-time
• 🧠 Neural Network Training: Actual TensorFlow models, not simulations
• 💬 Text Classification: Language detection and sentiment analysis
• 📊 Live Predictions: Interactive prediction with confidence scores
• 🎯 Multiple Model Types: MNIST, language detection, sentiment analysis

Quick Interactive Demo

./start-repl.sh

modelscript> load language_classifier
modelscript> predict
Enter text to classify> Bonjour comment allez vous
🎯 Prediction: French (87% confidence)

Enter text to classify> Hello how are you today  
🎯 Prediction: English (92% confidence)

---

🏗️ Project Structure

ModelScript/
├── modelscript/          # Core language engine
├── examples/             # Model definitions (.ms files)
│   ├── language_classifier.ms
│   ├── sentiment_classifier.ms
│   └── mnist_classifier.ms
├── data/                 # Training datasets
│   ├── language_data.csv
│   └── sentiment_data.csv
├── generated/            # Generated Python code
├── scripts/              # Utility scripts
├── repl.py               # Interactive REPL launcher
└── start-repl.sh         # Easy startup script

---

🚀 Quick Start

1. Setup Environment

# Clone and navigate
git clone https://github.com/rsbryan/modelscript.git
cd modelscript

# Create virtual environment
python3 -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate

# Install dependencies
pip install -r modelscript/requirements.txt

2. Launch Interactive REPL

./start-repl.sh

3. Try Interactive Examples

Language Detection:

modelscript> load language_classifier
modelscript> predict
# Test with French, Spanish, or English text!

Sentiment Analysis:

modelscript> load sentiment_classifier
modelscript> predict
# Test with positive or negative text!

View Model Structure:

modelscript> show
modelscript> generate  # See generated Python code

---

🧠 Interactive Features

Available Commands

Command	Description
examples	Show available model examples
load <model>	Load a model definition
show	Display current model structure
predict	🔥 Interactive prediction with neural training
generate	Show generated Python code
run	Train and execute model
help	Show all commands
quit	Exit REPL

Model Types

1. Language Classifier (language_classifier)

• Detects English, French, Spanish
• Uses LSTM neural network
• Trained on real multilingual data
• Interactive text classification

2. Sentiment Analyzer (sentiment_classifier)

• Positive/negative sentiment detection
• LSTM-based architecture
• Real-time confidence scoring
• Interactive text analysis

3. MNIST Classifier (mnist_classifier)

• Handwritten digit recognition
• Dense neural network
• Standard computer vision benchmark

---

💻 Traditional Usage (Non-Interactive)

Generate Python Code

python3 modelscript/modelscript.py examples/language_classifier.ms -o my_model.py

Run Model Directly

python3 modelscript/modelscript.py examples/sentiment_classifier.ms --run

---

📝 ModelScript Syntax

Basic Model Structure

model MyModel {
  dataset {
    source: "language"           // Built-in: "mnist", "language", "sentiment"
    max_sequence_length: 100
    vocab_size: 10000
    num_classes: 3
  }

  architecture {
    layer Embedding {
      input_dim: 10000
      output_dim: 128
      input_length: 100
    }
    
    layer LSTM {
      units: 64
    }
    
    layer Dense {
      units: 3
      activation: "softmax"
    }
  }

  training {
    batch_size: 32
    epochs: 5
    optimizer: "adam"
    loss: "sparse_categorical_crossentropy"
    metrics: ["accuracy"]
  }
}

Supported Datasets

Dataset	Description	Use Case
"mnist"	Handwritten digits	Image classification
"language"	Multilingual text samples	Language detection
"sentiment"	Positive/negative text	Sentiment analysis
"custom"	Your own data	Custom classification

Supported Layers

• Text Processing: Embedding, LSTM, Bidirectional
• Core Layers: Dense, Dropout, Flatten
• CNN Layers: Conv2D, MaxPooling2D
• Activation: "relu", "softmax", "sigmoid"

---

🔥 Neural Network Features

Real Training Capabilities

• Actual TensorFlow Models: Not simulations - real neural networks
• Live Data Processing: Tokenization, preprocessing, training
• Model Persistence: Trained models saved for reuse
• Confidence Scoring: Prediction probabilities
• Fallback Systems: Rule-based backup if training fails

Training Data

Language Detection (data/language_data.csv):

• 30 text samples in English, French, Spanish
• Labeled for supervised learning
• Preprocessed automatically

Sentiment Analysis (data/sentiment_data.csv):

• 30 positive and negative text samples
• Binary classification dataset
• Real-world sentiment examples

---

🧪 Testing & Validation

Run Integration Test

python3 integration_test.py

Expected output:

🚀 ModelScript Integration Test
📊 Results: 5/5 tests passed
🎉 All tests passed! ModelScript upgrade successful!

Manual Testing

# Test basic functionality
python3 test_interactive.py

# Test model parsing
python3 modelscript/modelscript/modelscript.py examples/language_classifier.ms

---

🛠️ Development

Adding New Models

1. Create a .ms file in examples/
2. Add training data to data/ (if needed)
3. Test with load <your_model> in REPL

Extending Datasets

Add new dataset support in modelscript/modelscript.py:

• Update generate_python_code() function
• Add data loading logic
• Update REPL prediction methods

---

🎯 Examples Gallery

Language Detection Example

modelscript> load language_classifier
modelscript> predict

Enter text to classify> Hola como estas
🎯 Prediction: Spanish (78% confidence)

Enter text to classify> Je suis très heureux
🎯 Prediction: French (91% confidence)

Sentiment Analysis Example

modelscript> load sentiment_classifier  
modelscript> predict

Enter text to analyze> This movie is absolutely amazing!
🎯 Prediction: 😊 Positive (94% confidence)

Enter text to analyze> This is the worst experience ever
🎯 Prediction: 😞 Negative (89% confidence)

---

📚 Advanced Usage

Custom Training Data

Replace data files with your own:

• data/language_data.csv - For language models
• data/sentiment_data.csv - For sentiment models

Format: text,label with proper headers

Generated Code Output

View the generated TensorFlow code:

modelscript> generate

This shows the actual Python/TensorFlow code that ModelScript creates from your .ms definition.

---

🔧 Troubleshooting

Virtual Environment Issues

# Recreate virtual environment
rm -rf venv
python3 -m venv venv
source venv/bin/activate
pip install -r modelscript/requirements.txt

Permission Issues

chmod +x start-repl.sh

Import Errors

Ensure you're in the correct directory and virtual environment is activated.

---

📄 License

MIT License - see LICENSE file for details.

---

🚀 What You Built

This is now a complete interactive ML development platform that:

✅ Parses declarative model definitions
✅ Generates executable TensorFlow/Keras code
✅ Trains real neural networks interactively
✅ Provides live text classification with confidence
✅ Supports multiple model types and datasets
✅ Includes comprehensive examples and testing

From simple DSL to production-ready interactive ML environment!