🧠 Eregion - Neural Network Analytics & Monitoring

Real-time neural network monitoring and analytics for PyTorch and TensorFlow

🚀 What is Eregion?

Eregion is a powerful Python library that provides real-time monitoring and analytics for your neural networks. Whether you're training models with PyTorch or TensorFlow, Eregion gives you deep insights into your model's behavior, layer activations, gradients, and performance metrics.

✨ Key Features

• 🔍 Layer-by-layer monitoring - Track activations, gradients, and outputs from every layer
• 📊 Advanced analytics - Dead neuron detection, entropy analysis, gradient norms
• ☁️ Cloud integration - Push metrics to eregion.dev for remote monitoring
• 🔄 Auto-tracking - Automatically monitor your model during training
• 🎯 Framework agnostic - Works seamlessly with PyTorch and TensorFlow
• 📈 Real-time metrics - Get instant feedback on model performance

🛠️ Installation

pip install eregion

Or install from source:

git clone https://github.com/BitLegion/eregion.git
cd eregion
pip install -e .

🚀 Quick Start

PyTorch Example

import torch
import torch.nn as nn
from eregion import EregionPyTorch

# Your model
model = nn.Sequential(
    nn.Linear(784, 128),
    nn.ReLU(),
    nn.Linear(128, 10)
)

# Initialize Eregion monitoring
tracker = EregionPyTorch(
    model=model,
    name="my_pytorch_model",
    API_KEY="your_api_key_here",
    auto_track=True  # Automatically track all layers
)

# Train your model
optimizer = torch.optim.Adam(model.parameters())
criterion = nn.CrossEntropyLoss()

for epoch in range(10):
    for batch_x, batch_y in dataloader:
        optimizer.zero_grad()
        outputs = model(batch_x)
        loss = criterion(outputs, batch_y)
        loss.backward()
        optimizer.step()
        
        # Push metrics to cloud
        tracker.push({"loss": loss.item(), "epoch": epoch})

TensorFlow Example

import tensorflow as tf
from eregion import EregionTensorFlow

# Your model
model = tf.keras.Sequential([
    tf.keras.layers.Dense(128, activation='relu', input_shape=(784,)),
    tf.keras.layers.Dense(10, activation='softmax')
])

# Initialize Eregion monitoring
tracker = EregionTensorFlow(
    model=model,
    name="my_tensorflow_model",
    API_KEY="your_api_key_here",
    auto_track=True
)

# Train your model
model.compile(optimizer='adam', loss='sparse_categorical_crossentropy')

for epoch in range(10):
    history = model.fit(
        train_dataset,
        epochs=1,
        verbose=0
    )
    
    # Push metrics to cloud
    tracker.push({
        "loss": history.history['loss'][0],
        "epoch": epoch
    })

📊 Analytics Features

Layer Monitoring

• Activation tracking - Monitor output distributions from each layer
• Gradient analysis - Track gradient norms and vanishing/exploding gradients
• Shape monitoring - Automatic detection of tensor shapes and dimensions

Advanced Metrics

• Dead neuron detection - Identify neurons with zero or near-zero activations
• Entropy analysis - Measure information content in layer outputs
• Activation distribution - Analyze statistical properties of activations

Cloud Integration

• Real-time dashboard - View metrics on eregion.dev
• Historical tracking - Compare performance across training runs
• Team collaboration - Share model insights with your team

🔧 Configuration

API Setup

1. Sign up at eregion.dev
2. Get your API key from the dashboard
3. Use the key in your Eregion tracker initialization

Auto-tracking Options

# Enable automatic layer monitoring
tracker = EregionPyTorch(
    model=model,
    name="my_model",
    API_KEY="your_key",
    auto_track=True,  # Automatically monitor all layers
    reset=True        # Reset existing model data
)

Manual Data Pushing

# Push custom metrics
tracker.push({
    "custom_metric": 0.95,
    "training_step": 1000,
    "learning_rate": 0.001
})

# Push layer-specific data
tracker.push({
    "layer_name": "conv1",
    "activation_mean": 0.1,
    "activation_std": 0.05
})

📈 What You Can Monitor

Training Metrics

• Loss values and convergence
• Learning rate schedules
• Gradient norms and distributions
• Parameter updates

Model Health

• Dead neurons and saturation
• Layer activation patterns
• Weight distributions
• Model complexity metrics

Performance Insights

• Training vs validation metrics
• Overfitting detection
• Model efficiency analysis
• Resource utilization

Development Setup

git clone https://github.com/pratyaypandey/eregion.git](https://github.com/sohanreddyk/neural-network-analytics-and-monitoring
cd eregion
pip install -e ".[dev]"
pytest tests/