Dataset API

Dataset

class ray.data.Dataset(plan: ExecutionPlan, logical_plan: LogicalPlan)

A Dataset is a distributed data collection for data loading and processing.

Datasets are distributed pipelines that produce ObjectRef[Block] outputs, where each block holds data in Arrow format, representing a shard of the overall data collection. The block also determines the unit of parallelism. For more details, see Ray Data Internals.

Datasets can be created in multiple ways: from synthetic data via range_*() APIs, from existing memory data via from_*() APIs (this creates a subclass of Dataset called MaterializedDataset), or from external storage systems such as local disk, S3, HDFS etc. via the read_*() APIs. The (potentially processed) Dataset can be saved back to external storage systems via the write_*() APIs.

Examples

import ray
# Create dataset from synthetic data.
ds = ray.data.range(1000)
# Create dataset from in-memory data.
ds = ray.data.from_items(
    [{"col1": i, "col2": i * 2} for i in range(1000)]
)
# Create dataset from external storage system.
ds = ray.data.read_parquet("s3://bucket/path")
# Save dataset back to external storage system.
ds.write_csv("s3://bucket/output")

Dataset has two kinds of operations: transformation, which takes in Dataset and outputs a new Dataset (e.g. map_batches()); and consumption, which produces values (not a data stream) as output (e.g. iter_batches()).

Dataset transformations are lazy, with execution of the transformations being triggered by downstream consumption.

Dataset supports parallel processing at scale: transformations such as map_batches(), aggregations such as min()/max()/mean(), grouping via groupby(), shuffling operations such as sort(), random_shuffle(), and repartition().

Examples

>>> import ray
>>> ds = ray.data.range(1000)
>>> # Transform batches (Dict[str, np.ndarray]) with map_batches().
>>> ds.map_batches(lambda batch: {"id": batch["id"] * 2})  
MapBatches(<lambda>)
+- Dataset(num_rows=1000, schema={id: int64})
>>> # Compute the maximum.
>>> ds.max("id")
999
>>> # Shuffle this dataset randomly.
>>> ds.random_shuffle()  
RandomShuffle
+- Dataset(num_rows=1000, schema={id: int64})
>>> # Sort it back in order.
>>> ds.sort("id")  
Sort
+- Dataset(num_rows=1000, schema={id: int64})

Both unexecuted and materialized Datasets can be passed between Ray tasks and actors without incurring a copy. Dataset supports conversion to/from several more featureful dataframe libraries (e.g., Spark, Dask, Modin, MARS), and are also compatible with distributed TensorFlow / PyTorch.

Basic Transformations

Dataset.add_column	Add the given column to the dataset.
Dataset.drop_columns	Drop one or more columns from the dataset.
Dataset.filter	Filter out rows that don't satisfy the given predicate.
Dataset.flat_map	Apply the given function to each row and then flatten results.
Dataset.limit	Truncate the dataset to the first limit rows.
Dataset.map	Apply the given function to each row of this dataset.
Dataset.map_batches	Apply the given function to batches of data.
Dataset.random_sample	Returns a new Dataset containing a random fraction of the rows.
Dataset.rename_columns	Rename columns in the dataset.
Dataset.select_columns	Select one or more columns from the dataset.

Consuming Data

Dataset.iter_batches	Return an iterable over batches of data.
Dataset.iter_rows	Return an iterable over the rows in this dataset.
Dataset.iter_tf_batches	Return an iterable over batches of data represented as TensorFlow tensors.
Dataset.iter_torch_batches	Return an iterable over batches of data represented as Torch tensors.
Dataset.iterator	Return a DataIterator over this dataset.
Dataset.show	Print up to the given number of rows from the Dataset.
Dataset.take	Return up to limit rows from the Dataset.
Dataset.take_all	Return all of the rows in this Dataset.
Dataset.take_batch	Return up to batch_size rows from the Dataset in a batch.

Execution

Dataset.materialize

Execute and materialize this dataset into object store memory.

Grouped and Global aggregations

Dataset.aggregate	Aggregate values using one or more functions.
Dataset.groupby	Group rows of a Dataset according to a column.
Dataset.max	Return the maximum of one or more columns.
Dataset.mean	Compute the mean of one or more columns.
Dataset.min	Return the minimum of one or more columns.
Dataset.std	Compute the standard deviation of one or more columns.
Dataset.sum	Compute the sum of one or more columns.
Dataset.unique	List the unique elements in a given column.

I/O and Conversion

Dataset.to_dask	Convert this Dataset into a Dask DataFrame.
Dataset.to_mars	Convert this Dataset into a Mars DataFrame.
Dataset.to_modin	Convert this Dataset into a Modin DataFrame.
Dataset.to_pandas	Convert this Dataset to a single pandas DataFrame.
Dataset.to_spark	Convert this Dataset into a Spark DataFrame.
Dataset.to_tf	Return a TensorFlow Dataset over this Dataset.
Dataset.to_torch	Return a Torch IterableDataset over this Dataset.
Dataset.write_csv	Writes the Dataset to CSV files.
Dataset.write_images	Writes the Dataset to images.
Dataset.write_json	Writes the Dataset to JSON and JSONL files.
Dataset.write_mongo	Writes the Dataset to a MongoDB database.
Dataset.write_numpy	Writes a column of the Dataset to .npy files.
Dataset.write_parquet	Writes the Dataset to parquet files under the provided path.
Dataset.write_tfrecords	Write the Dataset to TFRecord files.
Dataset.write_webdataset	Writes the dataset to WebDataset files.

Inspecting Metadata

Dataset.columns	Returns the columns of this Dataset.
Dataset.count	Count the number of rows in the dataset.
Dataset.input_files	Return the list of input files for the dataset.
Dataset.num_blocks	Return the number of blocks of this Dataset.
Dataset.schema	Return the schema of the dataset.
Dataset.size_bytes	Return the in-memory size of the dataset.
Dataset.stats	Returns a string containing execution timing information.

Sorting, Shuffling and Repartitioning

Dataset.random_shuffle	Randomly shuffle the rows of this Dataset.
Dataset.randomize_block_order	Randomly shuffle the blocks of this Dataset.
Dataset.repartition	Repartition the Dataset into exactly this number of blocks.
Dataset.sort	Sort the dataset by the specified key column or key function.

Splitting and Merging datasets

Dataset.split	Materialize and split the dataset into n disjoint pieces.
Dataset.split_at_indices	Materialize and split the dataset at the given indices (like np.split).
Dataset.split_proportionately	Materialize and split the dataset using proportions.
Dataset.streaming_split	Returns n DataIterators that can be used to read disjoint subsets of the dataset in parallel.
Dataset.train_test_split	Materialize and split the dataset into train and test subsets.
Dataset.union	Concatenate Datasets across rows.
Dataset.zip	Zip the columns of this dataset with the columns of another.

Schema

class ray.data.Schema(base_schema: pyarrow.lib.Schema | PandasBlockSchema, *, data_context: DataContext | None = None)

Dataset schema.

base_schema

The underlying Arrow or Pandas schema.

PublicAPI (beta): This API is in beta and may change before becoming stable.

Developer API

Dataset.to_pandas_refs	Converts this Dataset into a distributed set of Pandas dataframes.
Dataset.to_numpy_refs	Converts this Dataset into a distributed set of NumPy ndarrays or dictionary of NumPy ndarrays.
Dataset.to_arrow_refs	Convert this Dataset into a distributed set of PyArrow tables.
Dataset.iter_internal_ref_bundles	Get an iterator over RefBundles belonging to this Dataset.
block.Block	alias of pyarrow.Table | pandas.DataFrame
block.BlockExecStats	Execution stats for this block.
block.BlockMetadata	Metadata about the block.
block.BlockAccessor	Provides accessor methods for a specific block.