pytorch-graph 0.2.5

Enhanced PyTorch neural network architecture visualization with flowchart diagrams. Install with 'pip install pytorch-graph', import with 'import pytorch_graph'

PyTorch Graph

Professional PyTorch neural network visualization toolkit with complete computational graph analysis. Transform your PyTorch models into publication-ready diagrams with comprehensive architecture visualization and computational graph tracking.

Documentation

• Complete Documentation - Full API reference, guides, and examples
• Quick Start Guide - Get started in minutes
• API Reference - Complete function and class documentation
• Examples - Comprehensive usage examples

Demo

Try the library immediately with our included demo:

# Clone the repository
git clone https://github.com/PluvioXO/pytorch-graph.git
cd pytorch-graph

# Run the demo
python3 demo.py

The demo generates:

• Architecture diagrams (flowchart + research paper styles)
• Computational graph visualization
• Model analysis with parameter counts and memory usage
• Professional quality output ready for publications

Output files are saved to demo_outputs/ directory.

Key Features

Architecture Visualization

• Professional Flowchart Diagrams: Clean, vertical flowchart visualization with enhanced styling
• Research Paper Style: Publication-ready diagrams with academic formatting
• Multiple Export Formats: High-quality PNG with customizable DPI (up to 300 DPI)
• Comprehensive Layer Analysis: Parameter counts, memory usage, and tensor shape transformations

Complete Computational Graph Analysis

• Maximal Graph Traversal: Captures the entire computational graph without artificial limits
• Full Operation Names: Displays complete method and object names (no truncation)
• Smart Arrow Positioning: Arrows connect node edges properly without crossing over boxes
• Compact Layout: Eliminates gaps and breaks for continuous graph flow
• Real-time Execution Tracking: Monitors forward/backward passes and tensor operations

Professional Quality

• Enhanced Color Schemes: Color-coded operation types and layer categories
• Intelligent Legend Positioning: Automatic legend placement without overlap
• Memory Analysis: Per-layer and total memory usage estimates
• Model Complexity Assessment: Automatic size classification (Small/Medium/Large)
• Data Flow Visualization: Tensor sizes and shapes displayed on connections

Installation

# Basic installation
pip install pytorch-graph

# With enhanced features
pip install pytorch-graph[full]

# Development version
pip install pytorch-graph[dev]

Quick Start

Architecture Visualization

import torch
import torch.nn as nn
from pytorch_graph import generate_architecture_diagram

# Define your model
model = nn.Sequential(
    nn.Linear(784, 128),
    nn.ReLU(),
    nn.Linear(128, 64),
    nn.ReLU(),
    nn.Linear(64, 10)
)

# Generate professional architecture diagram
generate_architecture_diagram(
    model=model,
    input_shape=(1, 784),
    output_path="model_architecture.png",
    title="Neural Network Architecture"
)

Complete Computational Graph Analysis

from pytorch_graph import ComputationalGraphTracker
import torch

# Create a tracker for your model
tracker = ComputationalGraphTracker(model)

# Start tracking
tracker.start_tracking()

# Run your model
input_tensor = torch.randn(1, 784)
output = model(input_tensor)

# Stop tracking and save complete graph
tracker.stop_tracking()
tracker.save_graph_png("complete_computational_graph.png")

Comprehensive Examples

CNN Architecture Visualization

# Convolutional Neural Network
cnn_model = nn.Sequential(
    nn.Conv2d(3, 32, 3, padding=1),
    nn.ReLU(),
    nn.MaxPool2d(2),
    nn.Conv2d(32, 64, 3, padding=1),
    nn.ReLU(),
    nn.MaxPool2d(2),
    nn.Flatten(),
    nn.Linear(64 * 8 * 8, 128),
    nn.ReLU(),
    nn.Linear(128, 10)
)

# Generate multiple diagram styles
generate_architecture_diagram(
    cnn_model, 
    input_shape=(1, 3, 32, 32),
    output_path="cnn_flowchart.png",
    style="flowchart"
)

generate_architecture_diagram(
    cnn_model, 
    input_shape=(1, 3, 32, 32),
    output_path="cnn_research.png",
    style="research_paper"
)

Advanced Computational Graph Analysis

from pytorch_graph import track_computational_graph, analyze_computational_graph

# Track complete computational graph
tracker = track_computational_graph(
    model=model,
    input_tensor=input_tensor,
    track_memory=True,
    track_timing=True,
    track_tensor_ops=True
)

# Save high-quality computational graph
tracker.save_graph_png(
    filepath="complete_graph.png",
    width=1600,
    height=1200,
    dpi=300,
    show_legend=True
)

# Analyze the computational graph
analysis = analyze_computational_graph(model, input_tensor, detailed=True)
print(f"Total operations: {analysis['summary']['total_nodes']}")
print(f"Execution time: {analysis['summary']['execution_time']:.4f}s")

Diagram Styles

Enhanced Flowchart (Default)

generate_architecture_diagram(model, input_shape, "flowchart.png", style="flowchart")

Features:

• Lightning bolt icons for activation functions
• Memory usage per layer (e.g., "~1.2MB")
• Data flow indicators on arrows
• Summary panel with total parameters and memory
• Color-coded model complexity

Research Paper Style

generate_architecture_diagram(model, input_shape, "paper.png", style="research_paper")

Features:

• Academic formatting and typography
• Clean, minimal design
• Publication-ready quality
• Professional color scheme

Standard Style

generate_architecture_diagram(model, input_shape, "standard.png", style="standard")

Features:

• Classic neural network visualization
• Balanced information density
• Traditional layout

Computational Graph Features

Complete Graph Traversal

• No Artificial Limits: Traverses entire autograd graph without depth/operation restrictions
• Cycle Detection: Prevents infinite recursion while capturing complete structure
• Full Operation Coverage: Shows every operation in the computational graph

Smart Visualization

• Full Method Names: Displays complete operation names without truncation
• Proper Arrow Connections: Arrows connect node edges without crossing over boxes
• Compact Layout: Eliminates empty depth levels for continuous flow
• Enhanced Spacing: Optimized node positioning and spacing

Professional Output

• High-Resolution Images: Up to 300 DPI for publication quality
• Intelligent Legends: Automatic positioning without overlap
• Color-Coded Operations: Different colors for different operation types
• Clean Typography: Professional fonts and text formatting

Model Analysis

from pytorch_graph import analyze_model

# Comprehensive model analysis
analysis = analyze_model(
    model=model,
    input_shape=(1, 784),
    detailed=True
)

print(f"Total Parameters: {analysis.get('total_params', 'N/A'):,}")
print(f"Trainable Parameters: {analysis.get('trainable_params', 'N/A'):,}")
print(f"Model Size: {analysis.get('model_size_mb', 'N/A'):.2f} MB")
print(f"Layer Count: {analysis.get('layer_count', 'N/A')}")

Advanced Configuration

Custom Computational Graph Settings

# Create tracker with custom settings
tracker = ComputationalGraphTracker(
    model=model,
    track_memory=True,      # Track memory usage
    track_timing=True,      # Track execution timing
    track_tensor_ops=True   # Track tensor operations
)

# Save with custom parameters
tracker.save_graph_png(
    filepath="custom_graph.png",
    width=2000,             # Custom width
    height=1500,            # Custom height
    dpi=300,                # High DPI
    show_legend=True,       # Show legend
    node_size=25,           # Node size
    font_size=12            # Font size
)

Architecture Diagram Customization

generate_architecture_diagram(
    model=model,
    input_shape=(1, 784),
    output_path="custom_architecture.png",
    title="Custom Model Architecture",
    style="flowchart",
    dpi=300,
    show_legend=True
)

Visual Examples

Architecture Diagrams

• Flowchart Style: Professional vertical flow with enhanced information
• Research Paper Style: Clean, academic formatting
• Standard Style: Traditional neural network visualization

Computational Graphs

• Complete Graph: Full autograd traversal without breaks
• Smart Layout: Compact positioning with proper arrow connections
• Full Names: Complete operation names without truncation

Performance Features

• Memory Tracking: Real-time memory usage monitoring
• Execution Timing: Performance analysis and timing
• Tensor Operations: Complete tensor operation tracking
• Optimized Rendering: Fast diagram generation
• Efficient Layout: Smart positioning algorithms

Requirements

• Python: ≥ 3.8
• PyTorch: ≥ 1.8.0
• matplotlib: ≥ 3.3.0
• numpy: ≥ 1.19.0

API Reference

Core Functions

• generate_architecture_diagram(): Create architecture diagrams
• track_computational_graph(): Track computational graph execution
• analyze_computational_graph(): Analyze graph structure and performance
• analyze_model(): Comprehensive model analysis

Classes

• ComputationalGraphTracker: Complete computational graph tracking
• GraphNode: Individual graph node representation
• GraphEdge: Graph edge representation

Contributing

We welcome contributions! Please see our Contributing Guidelines for details.

Development Setup

git clone https://github.com/your-username/pytorch-graph.git
cd pytorch-graph
pip install -e .[dev]

License

MIT License - see LICENSE file for details.

Acknowledgments

• Built for the PyTorch community
• Inspired by the need for better model visualization tools
• Designed for researchers, practitioners, and educators

Support

• Issues: GitHub Issues
• Discussions: GitHub Discussions
• Documentation: Full Documentation

---

PyTorch Graph - Professional PyTorch model visualization made simple, beautiful, and comprehensive.