4.1.10. Tensor¶
-
class
Tensor¶ Value with any dimensions.
Public Functions
-
bool
valid() const¶ Check whether the object is valid or not.
- Return
- true if the object is valid, false otherwise.
- Remark
- This returns false when the object is created through the default constructor or the object had been moved.
-
void
check_valid() const¶ Check whether the object is valid or not.
- Exceptions
primitiv::Error: This object is invalid.
-
Device &
device() const¶ Returns the Device object related to the internal memory.
- Return
- Device object.
-
float
to_float() const¶ Retrieves one internal value in the tensor.
- Return
- An internal float value.
- Remark
- This function can be used only when the tensor is a scalar and non-minibatched (i.e., shape() == Shape()).
-
std::vector<float>
to_vector() const¶ Retrieves internal values in the tensor as a vector.
- Return
- A list of the internal values.
- Remark
- Each resulting values a re ordered by the column-major order, and the batch size is assumed as the last dimension of the tensor.
-
std::vector<std::uint32_t>
argmax(std::uint32_t dim) const¶ Retrieves argmax indices along an axis.
- Return
- A list of integers that indicates positions of the maximum values.
- Parameters
dim: A specified axis.
-
std::vector<std::uint32_t>
argmin(std::uint32_t dim) const¶ Retrieves argmin indices along an axis.
- Return
- A list of integers that indicates positions of the minimum values.
- Parameters
dim: A specified axis.
-
void
invalidate()¶ Invalidates this object.
-
void
reset(float k)¶ Reset internal values using a constant.
- Parameters
k: A value to be used to initialize each element.
-
void
reset_by_array(const float *values)¶ Reset internal values using a vector.
- Remark
- Length of
valuesshould be equal toshape().size(). Each element should be ordered by the column-major order, and the batch size is assumed as the last dimension. - Parameters
values: Array of values to be used to initialize each element.
-
void
reset_by_vector(const std::vector<float> &values)¶ Reset internal values using a vector.
- Remark
values.size()should be equal toshape().size(). Each element should be ordered by the column-major order, and the batch size is assumed as the last dimension.- Parameters
values: List of values to be used to initialize each element.
-
Tensor
reshape(const Shape &new_shape) const¶ Returns a tensor which have the same values and different shape.
- Return
- A new tensor.
- Parameters
new_shape: New shape with batch size 1.
-
Tensor &
inplace_multiply_const(float k)¶ Directly multiplies a constant.
- Return
*this- Parameters
k: A constant to multiply.
-
bool