Each Dataset
object is backed by a PyArrow Table.
A Table can be loaded from either the disk (memory mapped) or in memory.
Several Table types are available, and they all inherit from table.Table.
Wraps a pyarrow Table by using composition. This is the base class for InMemoryTable, MemoryMappedTable and ConcatenationTable.
It implements all the basic attributes/methods of the pyarrow Table class except the Table transforms: slice, filter, flatten, combine_chunks, cast, add_column, append_column, remove_column, set_column, rename_columns and drop.
The implementation of these methods differs for the subclasses.
( *args **kwargs )
Perform validation checks. An exception is raised if validation fails.
By default only cheap validation checks are run. Pass full=True for thorough validation checks (potentially O(n)).
(
*args
**kwargs
)
→
bool
Parameters
bool
, defaults to False
) —
Whether schema metadata equality should be checked as well.
Returns
bool
Check if contents of two tables are equal.
(
*args
**kwargs
)
→
List[pyarrow.RecordBatch]
Convert Table to list of (contiguous) RecordBatch objects.
Convert the Table to a dict or OrderedDict.
(
*args
**kwargs
)
→
pandas.Series
or pandas.DataFrame
Parameters
MemoryPool
, defaults to None
) —
Arrow MemoryPool to use for allocations. Uses the default memory
pool is not passed.
bool
, defaults to False
) —
Encode string (UTF8) and binary types to pandas.Categorical.
list
, defaults to empty
) —
List of fields that should be returned as pandas.Categorical. Only
applies to table-like data structures.
bool
, defaults to False
) —
Raise an ArrowException if this function call would require copying
the underlying data.
bool
, defaults to False
) —
Cast integers with nulls to objects
bool
, defaults to True
) —
Cast dates to objects. If False, convert to datetime64[ns] dtype.
bool
, defaults to False
) —
Cast non-nanosecond timestamps (np.datetime64) to objects. This is
useful if you have timestamps that don’t fit in the normal date
range of nanosecond timestamps (1678 CE-2262 CE).
If False, all timestamps are converted to datetime64[ns] dtype.
bool
, defaults to True
) —
Whether to parallelize the conversion using multiple threads.
bool
, defaults to False
) —
Do not create multiple copies Python objects when created, to save
on memory use. Conversion will be slower.
bool
, defaults to False
) —
If True, do not use the ‘pandas’ metadata to reconstruct the
DataFrame index, if present
bool
, defaults to True
) —
For certain data types, a cast is needed in order to store the
data in a pandas DataFrame or Series (e.g. timestamps are always
stored as nanoseconds in pandas). This option controls whether it
is a safe cast or not.
bool
, defaults to False
) —
If True, generate one internal “block” for each column when
creating a pandas.DataFrame from a RecordBatch or Table. While this
can temporarily reduce memory note that various pandas operations
can trigger “consolidation” which may balloon memory use.
bool
, defaults to False
) —
EXPERIMENTAL: If True, attempt to deallocate the originating Arrow
memory while converting the Arrow object to pandas. If you use the
object after calling to_pandas with this option it will crash your
program.
function
, defaults to None
) —
A function mapping a pyarrow DataType to a pandas ExtensionDtype.
This can be used to override the default pandas type for conversion
of built-in pyarrow types or in absence of pandas_metadata in the
Table schema. The function receives a pyarrow DataType and is
expected to return a pandas ExtensionDtype or None
if the
default conversion should be used for that type. If you have
a dictionary mapping, you can pass dict.get
as function.
Returns
pandas.Series
or pandas.DataFrame
pandas.Series
or pandas.DataFrame
depending on type of object
Convert to a pandas-compatible NumPy array or DataFrame, as appropriate
(
*args
**kwargs
)
→
pyarrow.Field
Select a schema field by its column name or numeric index.
(
*args
**kwargs
)
→
pyarrow.ChunkedArray
Select a column by its column name, or numeric index.
Iterator over all columns in their numerical order.
Schema of the table and its columns.
List of all columns in numerical order.
Number of columns in this table.
Number of rows in this table.
Due to the definition of a table, all columns have the same number of rows.
Dimensions of the table: (#rows, #columns).
Total number of bytes consumed by the elements of the table.
The table is said in-memory when it is loaded into the user’s RAM.
Pickling it does copy all the data using memory. Its implementation is simple and uses the underlying pyarrow Table methods directly.
This is different from the MemoryMapped table, for which pickling doesn’t copy all the data in memory. For a MemoryMapped, unpickling instead reloads the table from the disk.
InMemoryTable must be used when data fit in memory, while MemoryMapped are reserved for data bigger than memory or when you want the memory footprint of your application to stay low.
( *args **kwargs )
Perform validation checks. An exception is raised if validation fails.
By default only cheap validation checks are run. Pass full=True for thorough validation checks (potentially O(n)).
(
*args
**kwargs
)
→
bool
Parameters
bool
, defaults to False
) —
Whether schema metadata equality should be checked as well.
Returns
bool
Check if contents of two tables are equal.
(
*args
**kwargs
)
→
List[pyarrow.RecordBatch]
Convert Table to list of (contiguous) RecordBatch objects.
Convert the Table to a dict or OrderedDict.
(
*args
**kwargs
)
→
pandas.Series
or pandas.DataFrame
Parameters
MemoryPool
, defaults to None
) —
Arrow MemoryPool to use for allocations. Uses the default memory
pool is not passed.
bool
, defaults to False
) —
Encode string (UTF8) and binary types to pandas.Categorical.
list
, defaults to empty
) —
List of fields that should be returned as pandas.Categorical. Only
applies to table-like data structures.
bool
, defaults to False
) —
Raise an ArrowException if this function call would require copying
the underlying data.
bool
, defaults to False
) —
Cast integers with nulls to objects
bool
, defaults to True
) —
Cast dates to objects. If False, convert to datetime64[ns] dtype.
bool
, defaults to False
) —
Cast non-nanosecond timestamps (np.datetime64) to objects. This is
useful if you have timestamps that don’t fit in the normal date
range of nanosecond timestamps (1678 CE-2262 CE).
If False, all timestamps are converted to datetime64[ns] dtype.
bool
, defaults to True
) —
Whether to parallelize the conversion using multiple threads.
bool
, defaults to False
) —
Do not create multiple copies Python objects when created, to save
on memory use. Conversion will be slower.
bool
, defaults to False
) —
If True, do not use the ‘pandas’ metadata to reconstruct the
DataFrame index, if present
bool
, defaults to True
) —
For certain data types, a cast is needed in order to store the
data in a pandas DataFrame or Series (e.g. timestamps are always
stored as nanoseconds in pandas). This option controls whether it
is a safe cast or not.
bool
, defaults to False
) —
If True, generate one internal “block” for each column when
creating a pandas.DataFrame from a RecordBatch or Table. While this
can temporarily reduce memory note that various pandas operations
can trigger “consolidation” which may balloon memory use.
bool
, defaults to False
) —
EXPERIMENTAL: If True, attempt to deallocate the originating Arrow
memory while converting the Arrow object to pandas. If you use the
object after calling to_pandas with this option it will crash your
program.
function
, defaults to None
) —
A function mapping a pyarrow DataType to a pandas ExtensionDtype.
This can be used to override the default pandas type for conversion
of built-in pyarrow types or in absence of pandas_metadata in the
Table schema. The function receives a pyarrow DataType and is
expected to return a pandas ExtensionDtype or None
if the
default conversion should be used for that type. If you have
a dictionary mapping, you can pass dict.get
as function.
Returns
pandas.Series
or pandas.DataFrame
pandas.Series
or pandas.DataFrame
depending on type of object
Convert to a pandas-compatible NumPy array or DataFrame, as appropriate
(
*args
**kwargs
)
→
pyarrow.Field
Select a schema field by its column name or numeric index.
(
*args
**kwargs
)
→
pyarrow.ChunkedArray
Select a column by its column name, or numeric index.
Iterator over all columns in their numerical order.
Schema of the table and its columns.
List of all columns in numerical order.
Number of columns in this table.
Number of rows in this table.
Due to the definition of a table, all columns have the same number of rows.
Dimensions of the table: (#rows, #columns).
Total number of bytes consumed by the elements of the table.
Names of the table’s columns
( offset = 0 length = None ) → datasets.table.Table
Compute zero-copy slice of this Table
Select records from a Table. See pyarrow.compute.filter for full usage.
( *args **kwargs ) → datasets.table.Table
Flatten this Table. Each column with a struct type is flattened into one column per struct field. Other columns are left unchanged.
( *args **kwargs ) → datasets.table.Table
Make a new table by combining the chunks this table has.
All the underlying chunks in the ChunkedArray of each column are concatenated into zero or one chunk.
( *args **kwargs ) → datasets.table.Table
Cast table values to another schema
( *args **kwargs ) → datasets.table.Table
EXPERIMENTAL: Create shallow copy of table by replacing schema key-value metadata with the indicated new metadata (which may be None, which deletes any existing metadata
( *args **kwargs ) → datasets.table.Table
Add column to Table at position.
A new table is returned with the column added, the original table object is left unchanged.
( *args **kwargs ) → datasets.table.Table
Append column at end of columns.
( *args **kwargs ) → datasets.table.Table
Parameters
Returns
New table without the column.
Create new Table with the indicated column removed.
( *args **kwargs ) → datasets.table.Table
Replace column in Table at position.
Create new table with columns renamed to provided names.
( *args **kwargs ) → datasets.table.Table
Parameters
Returns
New table without the columns.
Raises
KeyError
KeyError
— : if any of the passed columns name are not existing.Drop one or more columns and return a new table.
( *args **kwargs ) → datasets.table.Table
Parameters
pandas.DataFrame
) —
pyarrow.Schema
, optional) —
The expected schema of the Arrow Table. This can be used to
indicate the type of columns if we cannot infer it automatically.
If passed, the output will have exactly this schema. Columns
specified in the schema that are not found in the DataFrame columns
or its index will raise an error. Additional columns or index
levels in the DataFrame which are not specified in the schema will
be ignored.
bool
, optional) —
Whether to store the index as an additional column in the resulting
Table
. The default of None will store the index as a column,
except for RangeIndex which is stored as metadata only. Use
preserve_index=True
to force it to be stored as a column.
int
, defaults to None
(may use up to system CPU count threads)) —
If greater than 1, convert columns to Arrow in parallel using
indicated number of threads
List[str]
, optional) —
List of column to be converted. If None, use all columns.
bool
, defaults to True
) —
Check for overflows or other unsafe conversions
Returns
Convert pandas.DataFrame to an Arrow Table.
The column types in the resulting Arrow Table are inferred from the dtypes of the pandas.Series in the DataFrame. In the case of non-object Series, the NumPy dtype is translated to its Arrow equivalent. In the case of object, we need to guess the datatype by looking at the Python objects in this Series.
Be aware that Series of the object dtype don’t carry enough information to always lead to a meaningful Arrow type. In the case that we cannot infer a type, e.g. because the DataFrame is of length 0 or the Series only contains None/nan objects, the type is set to null. This behavior can be avoided by constructing an explicit schema and passing it to this function.
( *args **kwargs ) → datasets.table.Table
Parameters
List[Union[pyarrow.Array, pyarrow.ChunkedArray]]
) —
Equal-length arrays that should form the table.
List[str]
, optional) —
Names for the table columns. If not passed, schema must be passed
Schema
, defaults to None
) —
Schema for the created table. If not passed, names must be passed
Union[dict, Mapping]
, default None) —
Optional metadata for the schema (if inferred).
Returns
Construct a Table from Arrow arrays
( *args **kwargs ) → datasets.table.Table
Construct a Table from Arrow arrays or columns
( *args **kwargs ) → datasets.table.Table
Construct a Table from a sequence or iterator of Arrow RecordBatches.
( table: Table path: str replays: typing.Union[typing.List[typing.Tuple[str, tuple, dict]], NoneType] = None )
The table is said memory mapped when it doesn’t use the user’s RAM but loads the data from the disk instead.
Pickling it doesn’t copy the data into memory. Instead, only the path to the memory mapped arrow file is pickled, as well as the list of transforms to “replay” when reloading the table from the disk.
Its implementation requires to store an history of all the transforms that were applied to the underlying pyarrow Table, so that they can be “replayed” when reloading the Table from the disk.
This is different from the InMemoryTable table, for which pickling does copy all the data in memory.
InMemoryTable must be used when data fit in memory, while MemoryMapped are reserved for data bigger than memory or when you want the memory footprint of your application to stay low.
( *args **kwargs )
Perform validation checks. An exception is raised if validation fails.
By default only cheap validation checks are run. Pass full=True for thorough validation checks (potentially O(n)).
(
*args
**kwargs
)
→
bool
Parameters
bool
, defaults to False
) —
Whether schema metadata equality should be checked as well.
Returns
bool
Check if contents of two tables are equal.
(
*args
**kwargs
)
→
List[pyarrow.RecordBatch]
Convert Table to list of (contiguous) RecordBatch objects.
Convert the Table to a dict or OrderedDict.
(
*args
**kwargs
)
→
pandas.Series
or pandas.DataFrame
Parameters
MemoryPool
, defaults to None
) —
Arrow MemoryPool to use for allocations. Uses the default memory
pool is not passed.
bool
, defaults to False
) —
Encode string (UTF8) and binary types to pandas.Categorical.
list
, defaults to empty
) —
List of fields that should be returned as pandas.Categorical. Only
applies to table-like data structures.
bool
, defaults to False
) —
Raise an ArrowException if this function call would require copying
the underlying data.
bool
, defaults to False
) —
Cast integers with nulls to objects
bool
, defaults to True
) —
Cast dates to objects. If False, convert to datetime64[ns] dtype.
bool
, defaults to False
) —
Cast non-nanosecond timestamps (np.datetime64) to objects. This is
useful if you have timestamps that don’t fit in the normal date
range of nanosecond timestamps (1678 CE-2262 CE).
If False, all timestamps are converted to datetime64[ns] dtype.
bool
, defaults to True
) —
Whether to parallelize the conversion using multiple threads.
bool
, defaults to False
) —
Do not create multiple copies Python objects when created, to save
on memory use. Conversion will be slower.
bool
, defaults to False
) —
If True, do not use the ‘pandas’ metadata to reconstruct the
DataFrame index, if present
bool
, defaults to True
) —
For certain data types, a cast is needed in order to store the
data in a pandas DataFrame or Series (e.g. timestamps are always
stored as nanoseconds in pandas). This option controls whether it
is a safe cast or not.
bool
, defaults to False
) —
If True, generate one internal “block” for each column when
creating a pandas.DataFrame from a RecordBatch or Table. While this
can temporarily reduce memory note that various pandas operations
can trigger “consolidation” which may balloon memory use.
bool
, defaults to False
) —
EXPERIMENTAL: If True, attempt to deallocate the originating Arrow
memory while converting the Arrow object to pandas. If you use the
object after calling to_pandas with this option it will crash your
program.
function
, defaults to None
) —
A function mapping a pyarrow DataType to a pandas ExtensionDtype.
This can be used to override the default pandas type for conversion
of built-in pyarrow types or in absence of pandas_metadata in the
Table schema. The function receives a pyarrow DataType and is
expected to return a pandas ExtensionDtype or None
if the
default conversion should be used for that type. If you have
a dictionary mapping, you can pass dict.get
as function.
Returns
pandas.Series
or pandas.DataFrame
pandas.Series
or pandas.DataFrame
depending on type of object
Convert to a pandas-compatible NumPy array or DataFrame, as appropriate
(
*args
**kwargs
)
→
pyarrow.Field
Select a schema field by its column name or numeric index.
(
*args
**kwargs
)
→
pyarrow.ChunkedArray
Select a column by its column name, or numeric index.
Iterator over all columns in their numerical order.
Schema of the table and its columns.
List of all columns in numerical order.
Number of columns in this table.
Number of rows in this table.
Due to the definition of a table, all columns have the same number of rows.
Dimensions of the table: (#rows, #columns).
Total number of bytes consumed by the elements of the table.
Names of the table’s columns
( offset = 0 length = None ) → datasets.table.Table
Compute zero-copy slice of this Table
Select records from a Table. See pyarrow.compute.filter for full usage.
( *args **kwargs ) → datasets.table.Table
Flatten this Table. Each column with a struct type is flattened into one column per struct field. Other columns are left unchanged.
( *args **kwargs ) → datasets.table.Table
Make a new table by combining the chunks this table has.
All the underlying chunks in the ChunkedArray of each column are concatenated into zero or one chunk.
( *args **kwargs ) → datasets.table.Table
Cast table values to another schema
( *args **kwargs ) → datasets.table.Table
EXPERIMENTAL: Create shallow copy of table by replacing schema key-value metadata with the indicated new metadata (which may be None, which deletes any existing metadata
( *args **kwargs ) → datasets.table.Table
Add column to Table at position.
A new table is returned with the column added, the original table object is left unchanged.
( *args **kwargs ) → datasets.table.Table
Append column at end of columns.
( *args **kwargs ) → datasets.table.Table
Parameters
Returns
New table without the column.
Create new Table with the indicated column removed.
( *args **kwargs ) → datasets.table.Table
Replace column in Table at position.
Create new table with columns renamed to provided names.
( *args **kwargs ) → datasets.table.Table
Parameters
Returns
New table without the columns.
Raises
KeyError
KeyError
— : if any of the passed columns name are not existing.Drop one or more columns and return a new table.
( table: Table blocks: typing.List[typing.List[datasets.table.TableBlock]] )
The table comes from the concatenation of several tables called blocks. It enables concatenation on both axis 0 (append rows) and axis 1 (append columns).
The underlying tables are called “blocks” and can be either InMemoryTable or MemoryMappedTable objects. This allows to combine tables that come from memory or that are memory mapped. When a ConcatenationTable is pickled, then each block is pickled:
Its implementation requires to store each block separately.
The blocks
attributes stores a list of list of blocks.
The first axis concatenates the tables along the axis 0 (it appends rows),
while the second axis concatenates tables along the axis 1 (it appends columns).
If some columns are missing when concatenating on axis 0, they are filled with null values. This is done using pyarrow.concat_tables(tables, promote=True).
You can access the fully combined table by accessing the ConcatenationTable.table attribute, and the blocks by accessing the ConcatenationTable.blocks attribute.
( *args **kwargs )
Perform validation checks. An exception is raised if validation fails.
By default only cheap validation checks are run. Pass full=True for thorough validation checks (potentially O(n)).
(
*args
**kwargs
)
→
bool
Parameters
bool
, defaults to False
) —
Whether schema metadata equality should be checked as well.
Returns
bool
Check if contents of two tables are equal.
(
*args
**kwargs
)
→
List[pyarrow.RecordBatch]
Convert Table to list of (contiguous) RecordBatch objects.
Convert the Table to a dict or OrderedDict.
(
*args
**kwargs
)
→
pandas.Series
or pandas.DataFrame
Parameters
MemoryPool
, defaults to None
) —
Arrow MemoryPool to use for allocations. Uses the default memory
pool is not passed.
bool
, defaults to False
) —
Encode string (UTF8) and binary types to pandas.Categorical.
list
, defaults to empty
) —
List of fields that should be returned as pandas.Categorical. Only
applies to table-like data structures.
bool
, defaults to False
) —
Raise an ArrowException if this function call would require copying
the underlying data.
bool
, defaults to False
) —
Cast integers with nulls to objects
bool
, defaults to True
) —
Cast dates to objects. If False, convert to datetime64[ns] dtype.
bool
, defaults to False
) —
Cast non-nanosecond timestamps (np.datetime64) to objects. This is
useful if you have timestamps that don’t fit in the normal date
range of nanosecond timestamps (1678 CE-2262 CE).
If False, all timestamps are converted to datetime64[ns] dtype.
bool
, defaults to True
) —
Whether to parallelize the conversion using multiple threads.
bool
, defaults to False
) —
Do not create multiple copies Python objects when created, to save
on memory use. Conversion will be slower.
bool
, defaults to False
) —
If True, do not use the ‘pandas’ metadata to reconstruct the
DataFrame index, if present
bool
, defaults to True
) —
For certain data types, a cast is needed in order to store the
data in a pandas DataFrame or Series (e.g. timestamps are always
stored as nanoseconds in pandas). This option controls whether it
is a safe cast or not.
bool
, defaults to False
) —
If True, generate one internal “block” for each column when
creating a pandas.DataFrame from a RecordBatch or Table. While this
can temporarily reduce memory note that various pandas operations
can trigger “consolidation” which may balloon memory use.
bool
, defaults to False
) —
EXPERIMENTAL: If True, attempt to deallocate the originating Arrow
memory while converting the Arrow object to pandas. If you use the
object after calling to_pandas with this option it will crash your
program.
function
, defaults to None
) —
A function mapping a pyarrow DataType to a pandas ExtensionDtype.
This can be used to override the default pandas type for conversion
of built-in pyarrow types or in absence of pandas_metadata in the
Table schema. The function receives a pyarrow DataType and is
expected to return a pandas ExtensionDtype or None
if the
default conversion should be used for that type. If you have
a dictionary mapping, you can pass dict.get
as function.
Returns
pandas.Series
or pandas.DataFrame
pandas.Series
or pandas.DataFrame
depending on type of object
Convert to a pandas-compatible NumPy array or DataFrame, as appropriate
(
*args
**kwargs
)
→
pyarrow.Field
Select a schema field by its column name or numeric index.
(
*args
**kwargs
)
→
pyarrow.ChunkedArray
Select a column by its column name, or numeric index.
Iterator over all columns in their numerical order.
Schema of the table and its columns.
List of all columns in numerical order.
Number of columns in this table.
Number of rows in this table.
Due to the definition of a table, all columns have the same number of rows.
Dimensions of the table: (#rows, #columns).
Total number of bytes consumed by the elements of the table.
Names of the table’s columns
( offset = 0 length = None ) → datasets.table.Table
Compute zero-copy slice of this Table
Select records from a Table. See pyarrow.compute.filter for full usage.
( *args **kwargs ) → datasets.table.Table
Flatten this Table. Each column with a struct type is flattened into one column per struct field. Other columns are left unchanged.
( *args **kwargs ) → datasets.table.Table
Make a new table by combining the chunks this table has.
All the underlying chunks in the ChunkedArray of each column are concatenated into zero or one chunk.
( target_schema *args **kwargs ) → datasets.table.Table
Cast table values to another schema
( *args **kwargs ) → datasets.table.Table
EXPERIMENTAL: Create shallow copy of table by replacing schema key-value metadata with the indicated new metadata (which may be None, which deletes any existing metadata
( *args **kwargs ) → datasets.table.Table
Add column to Table at position.
A new table is returned with the column added, the original table object is left unchanged.
( *args **kwargs ) → datasets.table.Table
Append column at end of columns.
( i *args **kwargs ) → datasets.table.Table
Parameters
Returns
New table without the column.
Create new Table with the indicated column removed.
( *args **kwargs ) → datasets.table.Table
Replace column in Table at position.
Create new table with columns renamed to provided names.
( columns *args **kwargs ) → datasets.table.Table
Parameters
Returns
New table without the columns.
Raises
KeyError
KeyError
— : if any of the passed columns name are not existing.Drop one or more columns and return a new table.
( tables: typing.List[typing.Union[pyarrow.lib.Table, datasets.table.Table]] axis: int = 0 )
Create ConcatenationTable from list of tables.
( tables: typing.List[datasets.table.Table] axis: int = 0 ) → datasets.table.Table
Parameters
Table
) — List of tables to be concatenated.
{0, 1}
, defaults to 0
, meaning over rows) —
Axis to concatenate over, where 0
means over rows (vertically) and 1
means over columns
(horizontally).
New in version 1.6.0
Returns
If the number of input tables is > 1, then the returned table is a datasets.table.ConcatenationTable
.
Otherwise if there’s only one table, it is returned as is.
Concatenate tables.
(
table: Table
)
→
List[str]
Returns
List[str]
a list of paths to the cache files loaded by the table
Get the cache files that are loaded by the table. Cache file are used when parts of the table come from the disk via memory mapping.