torcharrow

The torcharrow package contains data structures for two-dimensional, potentially heterogeneous tabular data, denoted as dataframe. It also defines relational operations over these dataframes. Additionally, it provides utilities for conversion with other formats (especially zero-copy conversion with Arrow arrays), and other useful utilities.

Creation and Conversion Ops

column([data, dtype, device])	Creates a TorchArrow Column.
dataframe([data, dtype, columns, device])	Creates a TorchArrow DataFrame.
from_arrow(data[, dtype, device])	Convert arrow array/table to a TorchArrow Column/DataFrame.
Column.to_arrow()	Convert self to arrow array

Mutating Ops

concat(columns)	Returns concatenated columns.
if_else(cond, left, right)	Return a column of elements where each of them is selected from either left column or righ column, depending on the value in the corresponding position in cond column.