arithmetic
Defines an arithmetic operator to the pipeline. The operator allows you to define complex operations between tensors. Different from the above overload which is defined by a string expression, this overload allows you to define the arithmetic operations directly, hence providing operand validations and type checks.
Parameters
the tensor to store the result of the arithmetic operation.
the arithmetic operations you would like apply onto Tensors. The last line of this closure will be taken as the result and assigned to result tensor. Note there are several requirements on the operand tensors involved in the arithmetic operations defined in this closure:
The arithmetic operations defined in one closure must not involve more than 10 Tensors. Please split into multiple arithmetic calls if you need to do so.
All the tensors involved in this closure must be matrices, i.e., must be multi-dimensional tensor (see Tensor.MultiDimensionalInitInfo) of exactly 2 dimensions.
All the tensors involved in the closure must have floating-point datatype, and its datatype must not be multi-dimensional pixel types, e.g., Tensor.DataType.Image.RG_FLOAT is not allowed.
See also
Throws
If the usage is not allowed by the SpatialML run-time Framework, or if the framework encounters internal error and cannot perform the requested behavior.
Deprecated
Directly defining arithmetic expressions in String skips the validation of the compatibility and type of operand tensors. Consider using other overloads of Pipeline.arithmetic to define the arithmetic expressions.
Replace with
arithmetic(result){ /*Expression*/ }Defines an arithmetic operator to the pipeline. An arithmetic operator takes in an array of PipelineTensor (size of which must be no more than 10), and write the arithmetic result into result. The arithmetic expression to be executed is defined by expression.
Note It is not recommended to use this method. Instead, please consider use the other overloads, because this method does not provide validation nor type checks of the operand tensors involved in the arithmetic expression. If you are confident to use it, you will need to explicitly OptIn.
@OptIn(AvoidArithmeticExpressionAsString::class)
pipeline.arithmetic("{0} + {1} * {3}", ...)Parameters
The arithmetic expression to be executed. It has the following requirements:
Use {IDX} to refer to the No. IDX input tensor, where IDX is the index among the operands parameter.
The following arithmetic symbols are allowed:
+,-,*(matrix multiplication),/(elementwise division),^(power, the right operand to which must be a constant or 1x1 tensor),sin,cos,tan,asin,acos,atan,sinh,cosh,tanhandinv( matrix inverse, for matrix division). Parentheses(and)are also supported.Constants are allowed in the expression, but they cannot appear on both sides of the arithmetic symbols listed above, such as
5.0 + 1.0is not valid, because the callers are required to pre-compute these arithmetic operations between constants. i.e., don't call the method with expression ={0} * (5.0 + 1.0), but{0} * 6.0instead.In terms of the computation priority ambiguity between trigonometry functions and power, we make the following rules:
sin {0} ^ 2means the sine of {0} ^ 2, whereas(sin {0}) ^ 2means the power 2 of sine of operand 0.
Valid examples are:
{0} + {1} * 2,sin (( {0} * 3.1415926 ) / 180),{0} * {1}note this is matrix multiplication,{0} * inv {1}.
the input operand to the arithmetic expression. The tensor indexed IDX in this array will replace the symbol {IDX} in expression. The size of this array must not exceed 10. All the tensors in this array must:
be created use Tensor.MultiDimensionalInitInfo as initialization config, i.e., must be multi-dimensional tensor with no special usage;
have exactly two dimensions (current limitation, we will support multi-dimensional arithmetic operations shortly afterwards).
the tensor to store the result of the arithmetic operation.
See also
Throws
If the usage is not allowed by the SpatialML run-time Framework, or if the framework encounters internal error and cannot perform the requested behavior.