Gunpowder Documentation

What is Gunpowder?

Gunpowder is a library to facilitate machine learning on large, multi-dimensional arrays.

Gunpowder allows you to assemble a pipeline from data loading over pre-processing, random batch sampling, data augmentation, pre-caching, training/prediction, to storage of results on arbitrarily large volumes of multi-dimensional images. Gunpowder is not tied to a particular learning framework, and thus complements libraries like PyTorch or TensorFlow.

• Overview
• Installation
• Tutorial: A Simple Pipeline
  • A minimal pipeline
  • Random samples
  • Geometric augmentation
  • Intensity augmentation
  • Creating batches with multiple samples
  • Requesting multiple arrays
  • Training a network
  • Predicting on a whole image
  • What next?
• Tutorial: Writing Your Own Node
  • The basics: prepare and process
  • Changing an array in-place
  • Requesting additional data
  • Creating new arrays
• API Reference
  • Data Containers
  • Requests and Specifications
  • Geometry
  • Node Base Classes
  • Source Nodes
  • Augmentation Nodes
  • Location Manipulation Nodes
  • Array Manipulation Nodes
  • Image Processing Nodes
  • Label Manipulation Nodes
  • Graph Processing Nodes
  • Provider Combination Nodes
  • Training and Prediction Nodes
  • Output Nodes
  • Performance Nodes
  • Iterative Processing Nodes

Indices and tables

• Index

• Module Index

• Search Page