cuDF User Guide

• API reference
  • Series
  • DataFrame
  • Index objects
  • GroupBy
  • General Functions
  • General Utilities
  • Window
  • Input/output
  • CharacterNormalizer
  • WordPieceTokenizer
  • TokenizeVocabulary
  • Options and settings
  • Extension Dtypes
  • Performance Tracking
• 10 Minutes to cuDF and Dask cuDF
  • What are these Libraries?
  • When to use cuDF and Dask cuDF
  • Object Creation
  • Viewing Data
  • Selecting a Column
  • Selecting Rows by Label
  • Selecting Rows by Position
  • Boolean Indexing
  • MultiIndex
  • Missing Data
  • Stats
  • Applymap
  • Histogramming
  • String Methods
  • Concat
  • Join
  • Grouping
  • Transpose
  • Time Series
  • Categoricals
  • Converting to Pandas
  • Converting to Numpy
  • Converting to Arrow
  • Reading/Writing CSV Files
  • Reading/Writing Parquet Files
  • Reading/Writing ORC Files
  • Dask Performance Tips
• Comparison of cuDF and Pandas
  • Supported operations
  • Data types
  • Null (or “missing”) values
  • Iteration
  • Result ordering
  • Floating-point computation
  • Column names
  • Writing a DataFrame to Parquet with non-string column names
  • No true "object" data type
  • .apply() function limitations
• Supported Data Types
  • NumPy data types
  • A note on object
  • Decimal data types
  • Nested data types (List and Struct)
• Input / Output
  • Input / Output
  • Working with JSON data
• Working with missing data
  • How to Detect missing values
  • Float dtypes and missing data
  • Datetimes
  • Calculations with missing data
  • Sum/product of Null/nans
  • NA values in GroupBy
  • Inserting missing data
  • Filling missing values: fillna
  • Filling with cudf Object
  • Dropping axis labels with missing data: dropna
  • Replacing generic values
  • String/regular expression replacement
  • Numeric replacement
• GroupBy
  • Summary of supported operations
  • Grouping
  • Aggregation
  • GroupBy apply
  • Transform
  • Rolling window calculations
• Overview of User Defined Functions with cuDF
  • Series UDFs
  • DataFrame UDFs
  • Rolling Window UDFs
  • GroupBy DataFrame UDFs
  • Numba Kernels on CuPy Arrays
  • Caveats
  • Summary
• Interoperability between cuDF and CuPy
  • Converting a cuDF DataFrame to a CuPy Array
  • Converting a cuDF Series to a CuPy Array
  • Converting a CuPy Array to a cuDF DataFrame
  • Converting a CuPy Array to a cuDF Series
  • Interweaving CuDF and CuPy for Smooth PyData Workflows
  • Converting a cuDF DataFrame to a CuPy Sparse Matrix
• Options
• Performance comparisons
  • Performance comparison
  • System Configuration
• Pandas Compatibility Notes
• Copy-on-write
  • Enabling copy-on-write
  • Disabling copy-on-write
  • Making copies
  • Notes
• Memory Profiling
  • Enabling Memory Profiling
• Breaking changes for pandas 2 in cuDF 24.04+
  • Removed DataFrame.append & Series.append, use cudf.concat instead.
  • Removed various numeric Index sub-classes, use cudf.Index
  • Change in bitwise operation results
  • ufuncs will perform re-indexing
  • DataFrame vs Series comparisons need to have matching index
  • Series.rank
  • Value counts sets the results name to count/proportion
  • DataFrame.describe will include datetime data by default
  • Converting a datetime string with Z to timezone-naive dtype is not allowed.
  • Datetime & Timedelta reduction operations will preserve their time resolutions.
  • get_dummies default return type is changed from int8 to bool
  • reset_index will name columns as None when name=None
  • Fixed an issue where duration components were being incorrectly calculated
  • fillna on datetime/timedelta with a lower-resolution scalar will now type-cast the series
  • Groupby.nth & Groupby.dtypes will have the grouped column in result