User guide

This user guide provides examples of using Awkward Array, organised by high-level topic into subsections. Each subsection provides an introduction to the topic, and is split into smaller pages that focus on a particular set of features.

New users should start with 10 minutes to Awkward Array.

• 10 minutes to Awkward Array
• Converting arrays
  • NumPy
  • Python objects
  • JSON
  • Arrow and Parquet
  • Pandas
  • Generic buffers
  • ROOT via Uproot
  • ROOT via RDataFrame
• Creating new arrays
  • Arrays of lists
  • Arrays of records
  • Arrays of missing data
  • Arrays of strings
  • Unflattening and grouping
  • ArrayBuilder (easy & general)
  • Direct constructors (fastest)
• Examining arrays
  • Data type
  • Single item detail [todo]
  • Listing fields/keys/columns
  • Simple slicing [todo]
  • Checking validity [todo]
• Numerical math
  • NumPy functions [todo]
  • Awkward broadcasting [todo]
  • Reducing (sum/min/any/all) [todo]
  • Statistics (mean/var/std) [todo]
  • Using argmin/argmax [todo]
  • On GPUs [todo]
• Filtering data
  • By number of items [todo]
  • Cuts vs. masks [todo]
  • Using ragged arrays
  • Using arrays with missing values
• Restructuring data
  • Zip/unzip and project
  • Adding fields to records
  • Renaming records [todo]
  • Flattening for plots
  • Padding/clipping for machine learning
  • Concatenating and interleaving [todo]
  • Sorting [todo]
• Combinatorics
  • Cartesian and "n choose k" [todo]
  • Best match between collections [todo]
• Using arrays in Numba
  • Supported features [todo]
  • Building array output [todo]
  • Working with CUDA
• Using arrays in C++
  • Building Awkward Arrays in C++ with LayoutBuilder
  • JIT compiling operations with C++ in cppyy
• Specialized behavior
  • Subclassing Array/Record [todo]
  • Overriding NumPy functions [todo]
  • In Numba [todo]
  • For physics: Lorentz vectors [todo]
  • Differentiation using JAX