stared/livelossplot

Live training loss plot in Jupyter Notebook for Keras, PyTorch and others

What it solves

livelossplot provides a way to visualize training progress in real-time within Jupyter Notebooks. It solves the problem of "training blindfolded," where developers rely on text logs to determine if a model is learning, stagnating, or overfitting.

How it works

The library provides a generic API (PlotLosses) and specialized callbacks for popular deep learning frameworks like Keras, PyTorch, Poutyne, and Ignite. During training, the library updates a plot with metrics (such as loss and accuracy) and renders it directly in the notebook cell output.

Who it’s for

Data scientists and researchers who use Jupyter Notebooks for exploration, teaching, or training small models and prefer a simple, plug-and-play alternative to complex tools like TensorBoard.

Highlights

  • Framework Agnostic: Supports Keras, PyTorch, and other frameworks via a bare API or dedicated callbacks.
  • Multiple Output Options: Supports rendering via Matplotlib and Bokeh.
  • Multiple Loggers: Includes loggers for standard output (ExtremaPrinter) and TensorBoard.
  • Jupyter Integration: Designed specifically for seamless live updates in Jupyter Notebook environments.

Related

  • Project
    intel/neural-compressor

    An open-source Python library that provides model compression and quantization techniques for PyTorch, TensorFlow, and JAX models to improve inference efficiency.

  • Project
    sktime/pytorch-forecasting

    A PyTorch-based library for time series forecasting that provides a high-level API and state-of-the-art deep learning architectures like Temporal Fusion Transformers and N-BEATS.

  • Project
    neuralhydrology/neuralhydrology

    A PyTorch-based Python library designed for deep learning research in hydrology, allowing users to train neural networks via configuration files.

  • Project
    sachinhosmani/torchvista

    An interactive visualization tool for PyTorch models that allows users to trace and visualize the forward pass directly in notebooks.

  • Project
    pytorch/pytorch

    A Python-based machine learning framework that provides GPU-accelerated tensor computation and a dynamic autograd system for building deep neural networks.