Rerun

TensorData

An N-dimensional array of numbers.

The number of dimensions and their respective lengths is specified by the shape field. The dimensions are ordered from outermost to innermost. For example, in the common case of a 2D RGB Image, the shape would be [height, width, channel].

These dimensions are combined with an index to look up values from the buffer field, which stores a contiguous array of typed values.

Fields fields

shape

Type: non-null List of UInt64

The shape of the tensor, i.e. the length of each dimension.

names

Type: List of Utf8

The names of the dimensions of the tensor (optional).

If set, should be the same length as datatypes.TensorData#shape. If it has a different length your names may show up improperly, and some constructors may produce a warning or even an error.

Example: ["height", "width", "channel", "batch"].

buffer

Type: non-null TensorBuffer

The content/data.

Arrow datatype arrow-datatype

Struct(
    "shape": non-null List(non-null UInt64)
    "names": List(non-null Utf8)
    "buffer": non-null Union(Dense,
        0: ("_null_markers": Null)
        1: ("U8": non-null List(non-null UInt8))
        2: ("U16": non-null List(non-null UInt16))
        3: ("U32": non-null List(non-null UInt32))
        4: ("U64": non-null List(non-null UInt64))
        5: ("I8": non-null List(non-null Int8))
        6: ("I16": non-null List(non-null Int16))
        7: ("I32": non-null List(non-null Int32))
        8: ("I64": non-null List(non-null Int64))
        9: ("F16": non-null List(non-null Float16))
        10: ("F32": non-null List(non-null Float32))
        11: ("F64": non-null List(non-null Float64))
    )
)

Used by used-by