rapidsai/cudf

Features

• GPU-Accelerated Data Analysis - Provides a high-level API for executing large-scale tabular filtering, joining, and aggregation directly on GPU hardware.
• Distributed GPU Computing - Integrates with Dask to scale tabular datasets across multiple GPU devices for memory-exceeding workloads.
• Data Import and Export - Parses CSV data into GPU dataframes and exports them with support for custom delimiters and encoding.
• Data Joins - Supports merging dataframes using various join types including inner, left, right, full, semi, and anti joins.
• Distributed Computing - Scales tabular datasets across multiple GPU devices to handle workloads that exceed the memory of a single device.
• GPU Data Processing Engines - Provides a high-level API for executing filtering, joining, and aggregating operations directly on GPU hardware.
• GPU DataFrame Libraries - Provides a GPU-accelerated library for manipulating and analyzing large tabular datasets.
• Grouped Aggregations - Performs GPU-accelerated grouped computations including mean, variance, standard deviation, and unique counts.
• High-Performance Data Analysis - Reads and writes Parquet, ORC, and CSV files directly to GPU memory to eliminate CPU processing bottlenecks.
• JSON Data Ingestion - Parses JSON and JSON Lines data into GPU dataframes with support for nested structures and schema inference.
• GPU Tabular Processing - Handles massive datasets using a GPU dataframe API to execute filtering, joins, and aggregations more efficiently.
• Lazy Query Execution - Executes lazy-evaluated Polars dataframe plans on a GPU engine to reduce execution time for complex computations.
• Memory Layouts - Implements optimized columnar memory layouts to maximize GPU bandwidth and SIMD execution efficiency.
• GPU Acceleration Backends - Intercepts standard dataframe calls to redirect computations to a GPU backend without requiring source code changes.
• Tabular Filtering - Enables extraction of rows and columns using boolean masks, indices, or logical expressions.
• Zero-Copy Data Access - Enables high-efficiency zero-copy data transfers between GPU dataframes and Apache Arrow or NumPy arrays.
• Zero-Copy Data Access Libraries - Integrates the Apache Arrow memory format to enable zero-copy data transfers between GPUs and other libraries.
• Python GPU Kernels - Compiles custom Python logic into high-performance CUDA kernels for GPU execution at runtime.
• Just-In-Time Compiled Functions - CuDF allows the compilation and execution of user-defined functions via just-in-time compilation for row-wise or grouped logic on the GPU.
• Just-in-Time Compilers - Includes a just-in-time compiler to execute custom user-defined functions for row-wise logic on the GPU.
• Sorting Algorithms - Arranges data in ascending or descending order based on one or more columns using stable sorting algorithms.
• Natural Language Processing - Implements GPU-accelerated linguistic tools such as tokenization, stemming, and edit distance calculations.
• GPU-Accelerated Text Processing - Provides GPU-accelerated regex, tokenization, and string normalization for large-scale natural language tasks.
• Streaming I/O - Copies data directly between GPU device buffers and supported filesystems via streaming I/O.
• ORC Parsers - Provides GPU-accelerated import and export functionality for ORC files, including stripe selection and column statistics.
• GPU-Accelerated Data Parsers - Provides a set of tools for reading and writing CSV, Parquet, and ORC files directly into GPU memory.
• GPUDirect Storage - Provides direct data streaming between GPU buffers and filesystems to eliminate CPU bottlenecks.
• Parquet Readers - Implements high-performance Parquet file loading into GPU memory with support for predicate pushdown.
• Predicate Pushdown - Filters data at the file level during Parquet or ORC ingestion to minimize GPU memory transfers.
• Temporal Data Processing - Provides time-based processing including timezone conversion, rounding, and calculation of calendrical sequences.
• Text String Manipulation Utilities - Provides GPU-accelerated regex replacement, splitting, and case normalization on large text columns.
• GPU Stream Scheduling - Assigns individual CUDA streams to host threads to overlap asynchronous data copies with kernel execution.
• Pinned Memory Pools - Manages reusable pools of page-locked host memory to accelerate data transfers between CPU and GPU.
• Rolling Aggregate Calculators - Calculates aggregations over sliding windows of data with support for grouped contexts.
• Abstract Syntax Tree Parsing - Parses logical expressions into an internal tree structure to execute complex boolean filtering on the GPU.
• Big Data and Distributed Computing - GPU-accelerated dataframe library.
• Data Manipulation - GPU-accelerated dataframe library.
• Specialized Data Utilities - GPU DataFrame library for data processing.
• Infrastructure and Deployment - GPU-accelerated data manipulation library.
• Managementul infrastructurii - Bibliotecă de manipulare a datelor accelerată GPU cu API similar pandas.
• Computation and Optimization - GPU-accelerated DataFrame library for data manipulation.
• Deep Learning Implementations - GPU-accelerated data manipulation library.

Istoric stele