Module org.apache.sis.referencing
Package org.apache.sis.parameter

Class TensorParameters<E>

Object
TensorParameters<E>
Type Parameters:
E - the type of tensor element values.
All Implemented Interfaces:
Serializable
public class TensorParameters<E> extends Object implements Serializable
Creates parameter groups for tensors (usually matrices). Matrices are handled as a special case of tensors (second-order tensors).

Each group of parameters contains the following elements:

  • Parameters (usually mandatory) for the tensor dimensions:
    • number of rows (named "num_row" in WKT1 conventions),
    • number of columns (named "num_col" in WKT1 conventions),
    • etc. for third-order or higher-order tensors.
  • A maximum of num_row × num_col × … optional parameters for the matrix or tensor element values. Parameter names depend on the formatting convention.
For all matrix or tensor elements, the default value is 1 for elements on the diagonal (where all indices have the same value) and 0 for all other elements. Those default values defines an identity matrix, or (more generally) Kroenecker delta tensor.

Parameters are not an efficient storage format for large tensors. Parameters are used only for small matrices/tensors to be specified in coordinate operations or processing libraries. In particular, those parameters integrate well in Well Known Text (WKT) format. For a more efficient matrix storage, see the matrix package.

Formatting

In the particular case of a tensor of rank 2 (i.e. a matrix), the parameters are typically formatted as below. Note that in the EPSG convention, the matrix is implicitly affine and of dimension 3×3.
Well Known Text (WKT) formats for matrix parameters
Using EPSG:9624 names and identifiers Using OGC names 
  Parameter["A0", <value>, Id["EPSG", 8623]],
  Parameter["A1", <value>, Id["EPSG", 8624]],
  Parameter["A2", <value>, Id["EPSG", 8625]],
  Parameter["B0", <value>, Id["EPSG", 8639]],
  Parameter["B1", <value>, Id["EPSG", 8640]],
  Parameter["B2", <value>, Id["EPSG", 8641]]
Note: the EPSG database also contains A3, A4, A5, A6, A7, A8 and B3 parameters, but they are for polynomial transformations, not for affine transformations.
 
  Parameter["num_row", 3],
  Parameter["num_col", 3],
  Parameter["elt_0_0", <value>],
  Parameter["elt_0_1", <value>],
  ...
  Parameter["elt_0_<num_col-1>", <value>],
  Parameter["elt_1_0", <value>],
  Parameter["elt_1_1", <value>],
  ...
  Parameter["elt_<num_row-1>_<num_col-1>", <value>]
Those groups are extensible, i.e. the number of "elt_row_col" parameters depends on the "num_row" and "num_col" parameter values. For this reason, the descriptor of matrix or tensor parameters is not immutable.

Usage examples

For creating a new group of parameters for a matrix using the WKT1 naming conventions, one can use the following code: 
Map<String,?> properties = Map.of(ParameterValueGroup.NAME_KEY, "Affine");
ParameterValueGroup p = TensorParameters.WKT1.createValueGroup(properties);
For setting the elements of a few values, then create a matrix from the parameter values: 
p.parameter("elt_0_0").setValue(4);    // "A0" also accepted as a synonymous of "elt_0_0".
p.parameter("elt_1_1").setValue(6);    // "B1" also accepted as a synonymous of "elt_1_1".
Matrix m = TensorParameters.WKT1.toMatrix(p);
Since:
0.4
See Also:

  • Field Details

    • ALPHANUM

      public static final TensorParameters<Double> ALPHANUM
      Parses and creates matrix parameters with alphanumeric names. Names are made of a letter indicating the row (first row is "A"), followed by a digit indicating the column index (first column is "0"). Aliases are the names as they were defined in version 1 of Well Known Text (WKT) format.
      Parameter names for a 3×3 matrix
      Primary name Alias 
         ┌            ┐
   │ A0  A1  A2 │
   │ B0  B1  B2 │
   │ C0  C1  C2 │
   └            ┘
      		 
         ┌                             ┐
   │ elt_0_0   elt_0_1   elt_0_2 │
   │ elt_1_0   elt_1_1   elt_1_2 │
   │ elt_2_0   elt_2_1   elt_2_2 │
   └                             ┘

      Relationship with EPSG

      The above-cited group of parameters are close, but not identical, to the definitions provided by the "Affine parametric transformation" (EPSG:9624) operation method. The differences are:
      • EPSG:9624 is for matrices of size 3×3 and does not provide any way to specify the matrix size. This ALPHANUM convention extends the definition to matrices of arbitrary size and accepts "num_row" and "num_col" as optional parameters.
      • EPSG:9624 is restricted to affine matrices and consequently define parameters only for the two first rows. This class accepts also parameters for the last row (namely "C0", "C1" and "C2" in a 3×3 matrices).
      Because of the above-cited extensions, this Tensor­Parameters constant cannot be named EPSG.
      Since:
      0.6

    • WKT1

      public static final TensorParameters<Double> WKT1
      Parses and creates matrix parameters with names matching the Well Known Text version 1 (WKT 1) convention.
      • First parameter is "num_row".
      • Second parameter is "num_col".
      • All other parameters are of the form "elt_row_col". Those parameters have alias of the form "A0", "A1", etc. where the letter indicates the row (first row is "A") and the digit is the column index (first column is "0").

      Example

      "elt_1_2" is the element name for the value at row 1 and column 2. Its alias is "B2", which is the EPSG name for the same parameter.

    • prefix

      protected final String prefix
      The prefix of parameter names for tensor elements. This is "elt_" in WKT 1.

    • separator

      protected final String separator
      The separator between row and column in parameter names for tensor elements. This is "_" in WKT 1.

  • Constructor Details

    • TensorParameters

      @SafeVarargs public TensorParameters(Class<E> elementType, String prefix, String separator, ParameterDescriptor<Integer>... dimensions)
      Constructs a descriptors provider.
      Parameters:
      element­Type - the type of tensor element values.
      prefix - the prefix to insert in front of parameter name for each tensor elements.
      separator - the separator between dimension (row, column, …) indices in parameter names.
      dimensions - the parameter for the size of each dimension, usually in an array of length 2. Length may be different if the caller wants to generalize usage of this class to tensors.

  • Method Details

    • getElementType

      public final Class<E> getElementType()
      Returns the type of tensor element values.
      Returns:
      the type of tensor element values.

    • rank

      public final int rank()
      Returns the rank of the tensor objects for which this instance will create parameters. The rank determines the type of objects represented by the parameters:
      Tensor types implied by rank
      Rank Type Used with
      0 scalar
      1 vector
      2 matrix Affine parametric transformation
      krank k tensor
      Returns:
      the rank of the tensors for which to create parameters.

    • getDimensionDescriptor

      public final ParameterDescriptor<Integer> getDimensionDescriptor(int i)
      Returns the parameter descriptor for the dimension at the given index.
      Parameters:
      i - the dimension index, from 0 inclusive to rank() exclusive.
      Returns:
      the parameter descriptor for the dimension at the given index.
      See Also:

    • getElementDescriptor

      public final ParameterDescriptor<E> getElementDescriptor(int... indices)
      Returns the parameter descriptor for a matrix or tensor element at the given indices. The length of the given indices array shall be equal to the rank. That length is usually 2, where indices[0] is the row index and indices[1] is the column index.
      Parameters:
      indices - the indices of the tensor element for which to get the descriptor.
      Returns:
      the parameter descriptor for the given tensor element.
      Throws:
      Illegal­Argument­Exception - if the given array does not have the expected length or have illegal value.
      See Also:

    • createElementDescriptor

      protected ParameterDescriptor<E> createElementDescriptor(int[] indices) throws IllegalArgumentException
      Creates a new parameter descriptor for a matrix or tensor element at the given indices. This method is invoked by get­Element­Descriptor(int[]) when a new descriptor needs to be created.

      Default implementation

      The default implementation converts the given indices to a parameter name by invoking the indices­To­Name(int[]) method, then creates a descriptor for an optional parameter of that name. The default value is given by get­Default­Value(int[]).

      Subclassing

      Subclasses can override this method if they want more control on descriptor properties like identification information, aliases or value domain.
      Parameters:
      indices - the indices of the tensor element for which to create a parameter.
      Returns:
      the parameter descriptor for the given tensor element.
      Throws:
      Illegal­Argument­Exception - if the given array does not have the expected length or have illegal value.
      See Also:

    • indicesToName

      protected String indicesToName(int[] indices) throws IllegalArgumentException
      Returns the parameter descriptor name of a matrix or tensor element at the given indices. The returned name shall be parsable by the name­To­Indices(String) method.

      Default implementation

      The default implementation requires an indices array having a length equals to the rank. That length is usually 2, where indices[0] is the row index and indices[1] is the column index. Then this method builds a name with the “prefix + row + separator + column + …” pattern (e.g. "elt_0_0").

      Subclassing

      If a subclass overrides this method for creating different names, then that subclass shall also override name­To­Indices(String) for parsing those names.
      Parameters:
      indices - the indices of the tensor element for which to create a parameter name.
      Returns:
      the parameter descriptor name for the tensor element at the given indices.
      Throws:
      Illegal­Argument­Exception - if the given array does not have the expected length or have illegal value.

    • nameToIndices

      protected int[] nameToIndices(String name) throws IllegalArgumentException
      Returns the indices of matrix element for the given parameter name, or null if none. This method is the converse of indices­To­Name(int[]).

      Default implementation

      The default implementation expects a name matching the “prefix + row + separator + column + …” pattern and returns an array containing the row, column and other indices, in that order.
      Parameters:
      name - the parameter name to parse.
      Returns:
      indices of the tensor element of the given name, or null if the name is not recognized.
      Throws:
      Illegal­Argument­Exception - if the name has been recognized but an error occurred while parsing it (e.g. an Number­Format­Exception, which is an Illegal­Argument­Exception subclass).

    • getDefaultValue

      protected E getDefaultValue(int[] indices)
      Returns the default value for the parameter descriptor at the given indices. The default implementation returns 1 if all indices are equals, or 0 otherwise.
      Parameters:
      indices - the indices of the tensor element for which to get the default value.
      Returns:
      the default value for the tensor element at the given indices, or null if none.
      Since:
      0.6
      See Also:

    • getAllDescriptors

      public ParameterDescriptor<?>[] getAllDescriptors(int... actualSize)
      Returns all parameters in this group for a tensor of the specified dimensions. The returned array contains all descriptors returned by get­Dimension­Descriptor(int) and get­Element­Descriptor(int...).
      Parameters:
      actual­Size - the matrix (or tensor) dimensions for which to get the parameters.
      Returns:
      the tensor parameters, including all elements.
      Since:
      0.6
      See Also:

    • createValueGroup

      public ParameterValueGroup createValueGroup(Map<String,?> properties)
      Creates a new instance of parameter group with default values of 1 on the diagonal, and 0 everywhere else. The returned parameter group is extensible, i.e. the number of elements will depend upon the value associated to the parameters that define the matrix (or tensor) dimension.

      The properties map is given unchanged to the identified object constructor. The following table is a reminder of main (not all) properties:

      Recognized properties (non exhaustive list)
      Property name Value type Returned by
      "name" ReferenceIdentifier or String AbstractIdentifiedObject.getName()
      "alias" GenericName or Char­Sequence (optionally as array) AbstractIdentifiedObject.getAlias()
      "identifiers" ReferenceIdentifier (optionally as array) AbstractIdentifiedObject.getIdentifiers()
      "remarks" InternationalString or String AbstractIdentifiedObject.getRemarks()
      Parameters:
      properties - the properties to be given to the identified object.
      Returns:
      a new parameter group initialized to the default values.

    • createValueGroup

      public ParameterValueGroup createValueGroup(Map<String,?> properties, Matrix matrix)
      Creates a new instance of parameter group initialized to the given matrix. This operation is allowed only for tensors of rank 2.
      Parameters:
      properties - the properties to be given to the identified object.
      matrix - the matrix to copy in the new parameter group.
      Returns:
      a new parameter group initialized to the given matrix.
      See Also:

    • toMatrix

      public Matrix toMatrix(ParameterValueGroup parameters) throws InvalidParameterNameException
      Constructs a matrix from a group of parameters. This operation is allowed only for tensors of rank 2.
      Parameters:
      parameters - the group of parameters.
      Returns:
      a matrix constructed from the specified group of parameters.
      Throws:
      Invalid­Parameter­Name­Exception - if a parameter name was not recognized.
      See Also:

    • hashCode

      public int hashCode()
      Returns a hash code value for this object.
      Overrides:
      hash­Code in class Object
      Returns:
      a hash code value.

    • equals

      public boolean equals(Object other)
      Compares this object with the given object for equality.
      Overrides:
      equals in class Object
      Parameters:
      other - the other object to compare with this object.
      Returns:
      true if both object are equal.