TFViz

Real-time visualization for TensorFlow training metrics with automatic color coding and professional plotting capabilities.

Features

• Real-time plotting with configurable update frequency
• Automatic metric categorization (loss, accuracy, precision, recall, etc.)
• Thread-safe operations for async plotting
• Multiple visualization themes (dark, light, minimal)
• TensorFlow Callback integration for seamless training monitoring
• Configurable plot styles (line, scatter, bar, histogram)
• Export/import capabilities for metric data
• Professional API with type hints and error handling
• Distinct color coding for training vs validation metrics

Quick Start

from tfviz import TFVizCallback

callback = TFVizCallback()
model.fit(X_train, y_train, callbacks=[callback])

Installation

pip install tfviz

Basic Usage

Simple Training Visualization

import tensorflow as tf
from tfviz import TFVizCallback

# Load your data
(x_train, y_train), (x_test, y_test) = tf.keras.datasets.mnist.load_data()

# Create your model
model = tf.keras.Sequential([
    tf.keras.layers.Dense(128, activation='relu'),
    tf.keras.layers.Dense(10, activation='softmax')
])
model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])

# Add TFViz callback
callback = TFVizCallback()

# Train with visualization
model.fit(x_train, y_train, epochs=10, validation_data=(x_test, y_test), callbacks=[callback])

Advanced Configuration

from tfviz import TFVizCallback, TFVizConfig
from tfviz.config import Theme, PlotStyle

# Custom configuration
config = TFVizConfig(
    theme=Theme.DARK,
    plot_config=PlotConfig(style=PlotStyle.LINE),
    update_config=UpdateConfig(frequency=5)
)

callback = TFVizCallback(config=config)
model.fit(X_train, y_train, callbacks=[callback])

Library Structure

tfviz/
├── __init__.py          # Main package exports
├── config.py            # Configuration classes and settings
├── metrics.py           # Metrics tracking and management
├── visualizer.py        # Visualization engine and plotting
└── callback.py          # TensorFlow callback integration

Components

TFVizConfig

Configuration management with themes, plot styles, and update settings.

MetricsTracker

Thread-safe metrics collection with automatic categorization and statistics.

VisualizationEngine

Matplotlib-based plotting engine with async capabilities and multiple themes.

TFVizCallback

TensorFlow callback for automatic metric collection and visualization during training.

Color Coding

The library automatically assigns distinct colors to different metrics:

• Training Loss: Red
• Validation Loss: Blue
• Training Accuracy: Green
• Validation Accuracy: Orange

Requirements

• Python 3.7+
• TensorFlow 2.x
• Matplotlib
• NumPy

License

MIT License