🚀 GPT-2 Pseudo-Code to Python Code Generator

📚 How to Use

1️⃣ Load Your Model

• Upload the best_model.pkl file (trained GPT-2 model)
• Click "Load Model" and wait for confirmation
• You'll see model configuration and training metrics

2️⃣ Generate Code

• Quick Start: Select an example from the dropdown
• Custom Input: Type your own pseudo-code description
• Optional: Add reference code to calculate BLEU scores
• Adjust generation parameters for different outputs
• Click "Generate Code"

3️⃣ Understand the Metrics

🎯 BLEU Score (Bilingual Evaluation Understudy)

• Measures similarity between generated and reference code
• BLEU-1: Word-level similarity (unigrams)
• BLEU-2: 2-word phrase similarity (bigrams)
• BLEU-3: 3-word phrase similarity (trigrams)
• BLEU-4: 4-word phrase similarity (most comprehensive)

Score Interpretation:

• 🟢 > 0.4: Excellent match - Generated code is very similar to reference
• 🟡 0.3-0.4: Good match - Code captures most key elements
• 🟠 0.2-0.3: Fair match - Some similarity exists
• 🔴 < 0.2: Poor match - Significant differences

📈 Additional Metrics

• Precision: How many generated words appear in reference
• Recall: How many reference words appear in generated code
• F1-Score: Harmonic mean of precision and recall
• Length Ratio: Generated vs reference code length
• Character Overlap: Character-level similarity

🎛️ Generation Parameters

💡 Example Pseudo-Code Prompts

Basic Operations

create a list of numbers from 1 to 10
define a function to calculate the sum of two numbers
iterate through a list and print each element

Conditionals & Logic

check if a number is even or odd
find the maximum of three numbers
validate if a string is empty

Data Structures

sort a list in descending order
remove duplicates from a list
merge two dictionaries

Algorithms

implement binary search algorithm
create a recursive function to calculate factorial
generate fibonacci sequence up to n terms
check if a string is palindrome

Advanced

create a class to represent a student with name and grades
implement a function to read CSV file and return dataframe
create a decorator to measure function execution time

🎓 About the Model

This model is fine-tuned on the SPOC (Search-based Pseudo-code to Code) dataset:

• 📄 Paper: SPOC: Search-based Pseudo-code to Code
• 🏛️ Source: Stanford University
• 🤖 Base Model: GPT-2 (Decoder-Only Transformer)
• 📊 Training: 10,000+ pseudo-code to code pairs
• 🎯 Task: Causal Language Modeling

⚠️ Limitations

• Model may not handle very complex algorithms perfectly
• Generated code should be tested before production use
• Best results with clear, specific pseudo-code descriptions
• Model trained on C++ code, adapted for Python generation

🤝 Tips for Best Results

1. ✅ Be Specific: "create a function to sort list in ascending order" vs "sort list"
2. ✅ Use Action Words: "create", "define", "implement", "calculate"
3. ✅ Mention Data Types: "list", "string", "dictionary", "integer"
4. ✅ Include Details: "recursive function" vs just "function"
5. ✅ Try Variations: Generate multiple times with different temperatures

📊 Generation History