Function pop

python/paddle/base/layers/math_op_patch.py:435–489

The type variable must be LoD Tensor Array. When self is DenseTensorArray, calling pop is similar to Python's pop on list. This interface is used to simplify dygraph to static graph operations. Args: self(Variable): The source variable, which must be DEN

(self, *args)

    @static_only
    def pop(self, *args):
        """
        The type variable must be LoD Tensor Array.
        When self is DenseTensorArray, calling pop is similar to Python's pop on list.
        This interface is used to simplify dygraph to static graph operations.

        Args:
            self(Variable): The source variable, which must be DENSE_TENSOR_ARRAY
            *args: optional, a int means index.
        Returns:
            Variable: self[index]
        """
        import paddle
        from paddle.static.nn import while_loop
        from paddle.tensor import fill_constant

        if self.type != core.VarDesc.VarType.DENSE_TENSOR_ARRAY:
            raise TypeError(
                f"Only Variable with VarType.DENSE_TENSOR_ARRAY support `pop` method, but received type: {self.type}"
            )
        if len(args) == 0:
            idx = -1
        else:
            idx = args[0]

        assert isinstance(idx, int)

        def cond(i, new_array):
            return paddle.less_than(i, arr_len)

        def body(i, new_array):
            item = paddle.tensor.array_read(array=self, i=i)
            paddle.tensor.array_write(
                item, paddle.tensor.array_length(new_array), new_array
            )

            i = paddle.increment(i)
            return i, new_array

        arr_len = paddle.tensor.array_length(self)
        if idx < 0:
            idx = idx + arr_len
        else:
            idx = fill_constant(shape=[1], dtype="int64", value=idx)

        pop_item = paddle.tensor.array_read(self, idx)

        tmp = paddle.assign(self)
        new_array = _slice_tensor_array(tmp, 0, idx)
        i = idx + 1

        _, new_array = while_loop(cond, body, [i, new_array])
        paddle.assign(new_array, output=self)

        return pop_item

    def _scalar_op_(var, scale, bias):
        block = current_block(var)