pyarrow.dataset.Dataset

class pyarrow.dataset.Dataset

Bases: pyarrow.lib._Weakrefable

Collection of data fragments and potentially child datasets.

Arrow Datasets allow you to query against data that has been split across multiple files. This sharding of data may indicate partitioning, which can accelerate queries that only touch some partitions (files).

__init__(*args, **kwargs)

Initialize self. See help(type(self)) for accurate signature.

Methods

__init__(*args, **kwargs)

Initialize self.

count_rows(self, **kwargs)

Count rows matching the scanner filter.

get_fragments(self, Expression filter=None)

Returns an iterator over the fragments in this dataset.

head(self, int num_rows, **kwargs)

Load the first N rows of the dataset.

replace_schema(self, Schema schema)

Return a copy of this Dataset with a different schema.

scanner(self, **kwargs)

Builds a scan operation against the dataset.

take(self, indices, **kwargs)

Select rows of data by index.

to_batches(self, **kwargs)

Read the dataset as materialized record batches.

to_table(self, **kwargs)

Read the dataset to an arrow table.

Attributes

partition_expression

An Expression which evaluates to true for all data viewed by this Dataset.

schema

The common schema of the full Dataset

count_rows(self, **kwargs)

Count rows matching the scanner filter.

See scanner method parameters documentation.

Returns

count (int)

get_fragments(self, Expression filter=None)

Returns an iterator over the fragments in this dataset.

Parameters

filter (Expression, default None) – Return fragments matching the optional filter, either using the partition_expression or internal information like Parquet’s statistics.

Returns

fragments (iterator of Fragment)

head(self, int num_rows, **kwargs)

Load the first N rows of the dataset.

See scanner method parameters documentation.

Returns

table (Table instance)

partition_expression

An Expression which evaluates to true for all data viewed by this Dataset.

replace_schema(self, Schema schema)

Return a copy of this Dataset with a different schema.

The copy will view the same Fragments. If the new schema is not compatible with the original dataset’s schema then an error will be raised.

scanner(self, **kwargs)

Builds a scan operation against the dataset.

Data is not loaded immediately. Instead, this produces a Scanner, which exposes further operations (e.g. loading all data as a table, counting rows).

Parameters
  • columns (list of str, default None) – The columns to project. This can be a list of column names to include (order and duplicates will be preserved), or a dictionary with {new_column_name: expression} values for more advanced projections. The columns will be passed down to Datasets and corresponding data fragments to avoid loading, copying, and deserializing columns that will not be required further down the compute chain. By default all of the available columns are projected. Raises an exception if any of the referenced column names does not exist in the dataset’s Schema.

  • filter (Expression, default None) – Scan will return only the rows matching the filter. If possible the predicate will be pushed down to exploit the partition information or internal metadata found in the data source, e.g. Parquet statistics. Otherwise filters the loaded RecordBatches before yielding them.

  • batch_size (int, default 1M) – The maximum row count for scanned record batches. If scanned record batches are overflowing memory then this method can be called to reduce their size.

  • use_threads (bool, default True) – If enabled, then maximum parallelism will be used determined by the number of available CPU cores.

  • use_async (bool, default False) – If enabled, an async scanner will be used that should offer better performance with high-latency/highly-parallel filesystems (e.g. S3)

  • memory_pool (MemoryPool, default None) – For memory allocations, if required. If not specified, uses the default pool.

  • fragment_scan_options (FragmentScanOptions, default None) – Options specific to a particular scan and fragment type, which can change between different scans of the same dataset.

Returns

scanner (Scanner)

Examples

>>> import pyarrow.dataset as ds
>>> dataset = ds.dataset("path/to/dataset")

Selecting a subset of the columns:

>>> dataset.scanner(columns=["A", "B"]).to_table()

Projecting selected columns using an expression:

>>> dataset.scanner(columns={
...     "A_int": ds.field("A").cast("int64"),
... }).to_table()

Filtering rows while scanning:

>>> dataset.scanner(filter=ds.field("A") > 0).to_table()
schema

The common schema of the full Dataset

take(self, indices, **kwargs)

Select rows of data by index.

See scanner method parameters documentation.

Returns

table (Table instance)

to_batches(self, **kwargs)

Read the dataset as materialized record batches.

See scanner method parameters documentation.

Returns

record_batches (iterator of RecordBatch)

to_table(self, **kwargs)

Read the dataset to an arrow table.

Note that this method reads all the selected data from the dataset into memory.

See scanner method parameters documentation.

Returns

table (Table instance)