Property M

property M: Tensor<bool> list -> Tensor<'T>

Picks elements from a tensor using one or more boolean mask tensors.

Declaration
property M: Tensor<bool> list -> Tensor<'T> with get, set
Parameters
Type Name Description
Tensor<bool> list masks

A list of boolean mask tensors or val NoMask.
Property Value
Type Description
Tensor<'T>

All elements from the tensor for which the mask is true.
Remarks

Masking picks elements from the tensor for which the corresponding element in the mask tensor is true. The mask can, for example, be generated by one or more element-wise comparison operation.

The get operation returns a copy of the selected elements of the tensor.

The set operation replaces the selected elements with a copy of the specified tensor.

If a dimension should not be masked, specify val NoMask instead of a mask tensor.

This mask specification variant is intended for programmatically generated ranges. For most use cases the variant property M: Tensor<bool> -> Tensor<'T> is more succinct and thus the recommended method.

Exceptions
Type Condition
System.InvalidArgumentException

Raised when the mask is incompatible with the tensor.
See Also

property M: Tensor<bool> -> Tensor<'T>

Picks elements from a tensor using one or more boolean mask tensors.

Declaration
property M: Tensor<bool> -> Tensor<'T> with get, set
Parameters
Type Name Description
Tensor<bool> m0

A boolean mask tensor or val NoMask.
Property Value
Type Description
Tensor<'T>

All elements from the tensor for which the mask is true.
Remarks

Masking picks elements from the tensor for which the corresponding element in the mask tensor is true. The mask can, for example, be generated by one or more element-wise comparison operation.

The get operation returns a copy of the selected elements of the tensor.

The set operation replaces the selected elements with a copy of the specified tensor.

If a dimension should not be masked, specify val NoMask instead of a mask tensor.

For clarity the documentation does not list all overloads of M. However, this masking method can be used for up to 5 dimensions, as shown in the example. For programmatically generated ranges or for more than 5 dimensions, the mask specification variant property M: Tensor<bool> list -> Tensor<'T> is available.

Currently this operation is only supported for tensors stored on the host. Support for CUDA tensors is planned in the future.

Examples
 let a = HostTensor.ofList [[1.0; 2.0; 3.0]
                            [4.0; 5.0; 6.0]]

 // masked get with one mask for the whole tensor
 let m = HostTensor.ofList [[true;  true;  false]
                            [false; false; true ]]
 let b = a.M(m) // b = [1.0; 2.0; 6.0]

 // an element-wise comparison operator can be used to create the mask
 let c = a.M(a >>>> 3.5) // c = [4.0; 5.0; 6.0]

 // masked get with one mask per dimension
 let m0 = HostTensor.ofList [true; false]
 let m1 = HostTensor.ofList [false; false; true]
 let d = a.M(m0, m1) // d = [3.0]

 // masked get using only one dimension
 let m0 = HostTensor.ofList [true; false]
 let e = a.M(m0, NoMask) // e = [[1.0; 2.0; 3.0]]

 // masked set with one mask for the whole tensor
 let m = HostTensor.ofList [[true;  true;  false]
                            [false; false; true ]]
 a.M(m) <- [8.0; 9.0; 0.0]  // a = [[8.0; 9.0; 3.0]
                            //      [4.0; 5.0; 0.0]]
Exceptions
Type Condition
System.InvalidArgumentException

Raised when the mask is incompatible with the tensor.
See Also