ai.rapids.cudf#
Interfaces#
A value that can be an operand of a binary operation, such as a column or a scalar. |
|
Provides a set of APIs for consuming host buffers. |
|
Provides a set of APIs for providing host buffers to be read. |
|
An allocator of host memory buffers. |
|
A resource that allocates/deallocates device memory. |
|
A callback interface for RMM memory allocation events. |
|
Provides an interface for reading multiple tables from a single input source. |
Classes#
Utility methods for breaking apart and reassembling 128-bit values during aggregations to enable hash-based aggregations and detect overflows. |
|
An Aggregation instance that also holds a column number and window metadata so the aggregation can be done over a specific window. |
|
Column builder from Arrow data. |
|
Options for reading data in Arrow IPC format. |
|
Settings for writing Arrow IPC data. |
|
This class is a Helper class to assert there are no non-empty nulls in a ColumnView. |
|
Options for reading an Avro file. |
|
Base class for all MemoryBuffers that are in device memory. |
|
Immutable byte range describing an offset and size within a file or buffer. |
|
Options for reading a CSV file. |
|
Options for writing a CSV file. |
|
JNI interface to cudf::chunked_pack. |
|
Utility class that wraps an array of closeable instances and can be closed. |
|
Base options class for input formats that can filter columns. |
|
This class represents the immutable vector of data. |
|
This class represents the column_view of a column analogous to its cudf cpp counterpart. |
|
Per column settings for writing Parquet/ORC files. |
|
Base writer options that add a compression type and key-value file metadata. |
|
Used to save groups and uniq key table for Table.contiguousSplitGroupsAndGenUniqKeys Each row in uniq key table is corresponding to a group Resource management note: This class is the owner of groups and uniqKeysTable(or uniqKeyColumns if table is not constructed) 1: Use closeGroups and closeUniqKeyTable to close the resources separately if you want to close eagerly. |
|
A table that is backed by a single contiguous device buffer. |
|
JNI wrapper for accessing the cuFile API. |
|
Represents a cuFile buffer. |
|
Represents a cuFile file handle for reading. |
|
Represents a cuFile file handle for reading. |
|
Java bindings for a subset of the CUDA runtime API. |
|
Represent free and total device memory. |
|
This class represents data allocated using cudaMalloc directly instead of the default RMM memory resource. |
|
Global settings for the native cuDF library. |
|
The data type of a column or scalar, with a scale for decimal types. |
|
Base class that can be used to provide data dynamically to CUDF. |
|
Utility methods for creating decimal types and comparing decimal columns. |
|
The default HostMemoryAllocator, which uses pinned memory when requested and available and falls back to off-heap memory. |
|
Provides JNI wrappers for reading Parquet files with deletion vector support. |
|
This class represents data in some form on the GPU. |
|
This class represents data in some form on the GPU. |
|
Byte range of a column chunk’s dictionary page, and whether that range is exactly the page. |
|
This class represents a reusable hash table built from distinct join keys from the right-side table for a join operation. |
|
A reusable hash lookup built from the right-side join keys for left semi and left anti joins. |
|
This class tracks the data associated with a gather map, a buffer of INT32 elements that index a source table and can be passed to a table gather operation. |
|
Options that control how ColumnView.getJSONObject(Scalar, GetJsonObjectOptions) parses JSON. |
|
An aggregation that can be used for a reduce. |
|
A GroupByAggregation for a specific column in a table. |
|
Options for groupby (see cudf::groupby::groupby’s constructor). |
|
An aggregation that can be used for a grouped scan. |
|
A GroupByScanAggregation for a specific column in a table. |
|
This class represents a hash table built from the join keys of the right-side table for a join operation. |
|
Similar to a ColumnVector, but the data is stored in host memory and accessible directly from the JVM. |
|
A class that holds Host side Column Vector APIs and the OffHeapState. |
|
This class holds an off-heap buffer in the host/CPU memory. |
|
A wrapper around native host UDF aggregations. |
|
Experimental Parquet hybrid-scan reader. |
|
Serialize and deserialize CUDF tables and columns using a custom format. |
|
Options for reading in JSON encoded data. |
|
Remaps keys to unique integer IDs. |
|
Abstract class for representing the Memory Buffer. |
|
ColumnVectors may store data off heap, and because of complicated processing the life time of an individual vector can vary a lot. |
|
This class tracks size information associated with a mixed table join. |
|
This is a DataSource that can take multiple HostMemoryBuffers. |
|
Command-line utilities for inspecting and extracting packaged native dependencies. |
|
This class will load the native dependencies. |
|
This class supports push/pop NVTX profiling ranges, or “scoped” ranges. |
|
This class supports start/end NVTX profiling ranges. |
|
Provide an interface for reading an ORC file in an iterative manner. |
|
Options for reading a ORC file. |
|
This class represents settings for writing ORC files. |
|
Provides the ordering for specific columns. |
|
Metadata for a table that is backed by a single contiguous device buffer. |
|
Provide an interface for reading a Parquet file in an iterative manner. |
|
Options for reading a parquet file. |
|
A chunked Parquet writer that can return its footer metadata when closed. |
|
This class represents settings for writing Parquet files. |
|
Class to provide a PartitionedTable. |
|
This is the JNI interface to an RMM pool backed by pinned host memory. |
|
Helper utility for creating ranges. |
|
An aggregation that can be used for a reduce. |
|
Regex program class, closely following cudf::strings::regex_program. |
|
A replacement policy for a specific column. |
|
This is the binding class for rmm lib. |
|
Bit flags that select the RMM memory allocation strategy. |
|
A device memory resource that will pre-allocate a pool of resources and sub-allocate from this pool to improve memory performance. |
|
A device memory resource that uses cudaMallocAsync and cudaFreeAsync for allocation and deallocation. |
|
A device memory resource that uses cudaMalloc and cudaFree for allocation and deallocation. |
|
A device memory resource that will give callbacks in specific situations. |
|
A device memory resource that will limit the maximum amount allocated. |
|
A device memory resource that will log interactions. |
|
A device memory resource that uses cudaMallocManaged and cudaFreeManaged for allocation and deallocation. |
|
A device memory resource that will pre-allocate a pool of resources and sub-allocate from this pool to improve memory performance. |
|
A device memory resource that will track some basic statistics about the memory usage. |
|
A resource that wraps another RmmDeviceMemoryResource. |
|
An aggregation that can be used on rolling windows. |
|
A RollingAggregation for a specific column in a table. |
|
A single scalar value. |
|
An aggregation that can be used for a scan. |
|
The schema of data to be read in. |
|
An aggregation that can be used for a reduce. |
|
Class to represent a collection of ColumnVectors and operations that can be performed on them collectively. |
|
A debugging utility that prints the contents of tables and columns. |
|
A table along with some metadata about the table. |
|
Provides an interface for writing out Table information in multiple steps. |
|
Owning result of a status-aware Variant extraction or cast. |
|
Utility methods for cuDF’s experimental Parquet Variant extraction APIs. |
|
Options for rolling windows. |
Enums#
Mathematical binary operations. |
|
Types of buffers supported by ColumnVectors and HostColumnVectors. |
|
Capture groups setting, closely following cudf::strings::capture_groups. |
|
Enumeration of compression formats. |
|
This is the Java mapping of CUDA device compute modes. |
|
Types of datetime components that may be extracted. |
|
The unit to which timestamps are rounded. |
|
Used for the dropListDuplicates function Specifies which duplicate to keep. |
|
Hash algorithm identifiers, mirroring native enum cudf::hash_id. |
|
Join semantics to apply when filtering equality-join gather maps of the corresponding kind. |
|
How should NaNs be compared in an operation. |
|
How should nulls be compared in an operation. |
|
Specify whether to include nulls or exclude nulls in an operation. |
|
The color of an NVTX range, as shown in profiling tools. |
|
Policy to account for possible out-of-bounds indices. |
|
The side of a string to which padding characters are added. |
|
Interpolation method to use when the desired quantile lies between two data points i and j. |
|
Quote style for CSV records, closely following cudf::io::quote_style. |
|
Regex flags setting, closely following cudf::strings::regex_flags. |
|
Policy to specify the position of replacement values relative to null rows. |
|
Scan operation type. |
|
Mathematical unary operations. |
|
Logical type identifiers returned by VariantUtils.getVariantTypeId(ColumnView). |
|
Per-row status identifiers returned by Variant extraction and cast operations. |
Exceptions#
Exception from the cuda language/library. |
|
CudaFatalException is a kind of CudaException which leaves the process in an inconsistent state and any further CUDA work will return the same error. |
|
Exception thrown when CUDF operation results in a column size exceeding CUDF column size limits. |
|
Exception thrown by cudf itself. |
|
Exception from RMM allocator. |
Annotation Types#
Marks a public type whose API is still considered experimental and may change without notice. |