Column object

class legate_dataframe.lib.core.column.LogicalColumn(obj: Any)

Logical column distributed between legate nodes

Underlying a logical column is a logical array. The column doesn’t own the array, a logical array can be part of multiple columns.

A column may represent columnar or scalar data. For a scalar column column.scalar() is True and it has always one row.

add_as_next_task_input(self, task: AutoTask) → None

Add a logical column to the next input task argument

This should match a call to get_next_input<PhysicalColumn>() by a legate task.

Note

The order of “add_next_*” calls must match the order of the corresponding “get_next_*” calls.

Parameters:

task – The legate task to add the argument.

add_as_next_task_output(self, task: AutoTask) → None

Add a logical column to the next output task argument

This should match a call to get_next_input<PhysicalColumn>() by a legate task.

Note

The order of “add_next_*” calls must match the order of the corresponding “get_next_*” calls.

Parameters:

argument. (The legate task to add the)

dtype(self) → DtypeObj

Returns the cudf data type of the row elements

Return type:

The cudf data type

static empty_like_logical_column(LogicalColumn col) → LogicalColumn

Create a new unbounded column from an existing column.

Parameters:

other (LogicalColumn) – The prototype column.

Return type:

The new unbounded column with the type and nullable equal other

static from_cudf(col_or_scalar) → LogicalColumn

Create a logical column from a local cudf column or column.

This call blocks the client’s control flow and scatter the data to all legate nodes. If the input is a DeviceScalar the column will be marked as scalar.

Parameters:

col – cudf column

Return type:

New logical column

get_logical_array(self, *, check_dtype=False) → LogicalArray

Return the underlying logical array

Parameters:

check_dtype – If True check that the dtype round-trips. Defaults to False.

Return type:

The underlying logical array

Raises:

TypeError – If check_dtype=True and the column dtype is not a native legate data type.

is_scalar(self)

num_rows(self) → int

Returns the number of rows.

Return type:

The number of rows

Raises:

RuntimeError – if column is unbound

offload_to(self, cpp_StoreTarget target_mem)

Offload the underlying data to the specified memory.

This method offloads the underlying data to the specified target memory. The purpose of this is to free up GPU memory resources. See legate::LogicalArray::offload_to() for more information.

Parameters:

target_mem (legate.core.StoreTarget) – The memory kind to offload to. To offload to the CPU use legate.core.StoreTarget.SYSMEM.

repr(self, size_t max_num_items=30) → str

Return a printable representational string

Parameters:

max_num_items (int) – Maximum number of items to include before items are abbreviated.

Return type:

Printable representational string

to_array(self, *, writeable=False)

Return a view into the column data.

This method returns a cupynumeric array viewing the columns data. For non-nullable columns cupynumeric.asarray(column) should also work directly.

Parameters:

writeable (bool) – By default sets the cupynumeric array not not writeable since columns are normally immutable. This can be overridden by passing True.

Returns:

A 1-d cupynumeric array.

Return type:

array

Raises:

• TypeError – If the dtype is unsupported by cupynumeric.

• ValueError – If the column has a mask and there are masked values.

to_cudf(self) → cudfColumn

Copy the logical column into a local cudf column

This call blocks the client’s control flow and fetches the data for the whole column to the current node.

Return type:

A cudf column that owns its data.

to_cudf_scalar(self) → DeviceScalar

Copy the logical column into a local cudf scalar

This call blocks the client’s control flow and fetches the data for the scalar. To succeed the column must have length one. Columns for which column.scalar() is True always have length 1.

Return type:

A cudf scalar that owns its data.

Raises:

ValueError – If the column is not length 1 (scalar columns always are).