MaskedLattice.h

Classes

MaskedLattice -- A templated, abstract base class for array-like objects with masks. (full description)

template <class T> class MaskedLattice : public Lattice<T>

Interface

Public Members

MaskedLattice() : itsDefRegPtr(0)

MaskedLattice (const MaskedLattice<T>&)

virtual ~MaskedLattice()

virtual MaskedLattice<T>* cloneML() const = 0

virtual Lattice<T>* clone() const

virtual Bool isMasked() const

virtual Bool hasPixelMask() const

virtual const Lattice<Bool>& pixelMask() const

virtual Lattice<Bool>& pixelMask()

const LatticeRegion& region() const

Bool getMask (COWPtr<Array<Bool> >& buffer, Bool removeDegenerateAxes=False) const

Bool getMaskSlice (COWPtr<Array<Bool> >& buffer, const Slicer& section, Bool removeDegenerateAxes=False) const

Bool getMaskSlice (COWPtr<Array<Bool> >& buffer, const IPosition& start, const IPosition& shape, Bool removeDegenerateAxes=False) const

Bool getMaskSlice (COWPtr<Array<Bool> >& buffer, const IPosition& start, const IPosition& shape, const IPosition& stride, Bool removeDegenerateAxes=False) const

Bool getMask (Array<Bool>& buffer, Bool removeDegenerateAxes=False)

Bool getMaskSlice (Array<Bool>& buffer, const Slicer& section, Bool removeDegenerateAxes=False)

Bool getMaskSlice (Array<Bool>& buffer, const IPosition& start, const IPosition& shape, Bool removeDegenerateAxes=False)

Bool getMaskSlice (Array<Bool>& buffer, const IPosition& start, const IPosition& shape, const IPosition& stride, Bool removeDegenerateAxes=False)

Array<Bool> getMask (Bool removeDegenerateAxes=False) const

Array<Bool> getMaskSlice (const Slicer& section, Bool removeDegenerateAxes=False) const

Array<Bool> getMaskSlice (const IPosition& start, const IPosition& shape, Bool removeDegenerateAxes=False) const

Array<Bool> getMaskSlice (const IPosition& start, const IPosition& shape, const IPosition& stride, Bool removeDegenerateAxes=False) const

virtual Bool doGetMaskSlice (Array<Bool>& buffer, const Slicer& section)

Protected Members

MaskedLattice<T>& operator= (const MaskedLattice<T>&)

virtual const LatticeRegion* getRegionPtr() const = 0

Description

Review Status

Date Reviewed:

yyyy/mm/dd

Programs:

Demos:

dLattice.cc

Prerequisite

IPosition
Abstract Base class Inheritance - try "Advanced C++" by James O. Coplien, Ch. 5.

Etymology

Lattice: "A regular, periodic configuration of points, particles, or objects, throughout an area of a space..." (American Heritage Directory) This definition matches our own: an n-dimensional arrangement of items, on regular orthogonal axes.

Synopsis

This pure abstract base class defines the operations which may be performed on any concrete class derived from it. It has only a few non-pure virtual member functions. The fundamental contribution of this class, therefore, is that it defines the operations derived classes must provide:

how to extract a "slice" (or sub-array, or subsection) from a Lattice.
how to copy a slice in.
how to get and put a single element
how to apply a function to all elements
various shape related functions.

Lattices are always zero origined.

Example

Because Lattice is an abstract base class, an actual instance of this class cannot be constructed. However the interface it defines can be used inside a function. This is always recommended as it allows Functions which have Lattices as arguments to work for any derived class.

I will give a few examples here and then refer the reader to the ArrayLattice class (a memory resident Lattice) and the PagedArray class (a disk based Lattice) which contain further examples with concrete classes (rather than an abstract one). All the examples shown below are used in the dLattice.cc demo program.

Example 1:

This example calculates the mean of the Lattice. Because Lattices can be too large to fit into physical memory it is not good enough to simply use getSlice to read all the elements into an Array. Instead the Lattice is accessed in chunks which can fit into memory (the size is determined by the maxPixels and niceCursorShape functions). The LatticeIterator::cursor() function then returns each of these chunks as an Array and the standard Array based functions are used to calculate the mean on each of these chunks. Functions like this one are the recommended way to access Lattices as the LatticeIterator will correctly setup any required caches.

    Complex latMean(const Lattice<Complex>& lat) {
      const uInt cursorSize = lat.advisedMaxPixels();
      const IPosition cursorShape = lat.niceCursorShape(cursorSize);
      const IPosition latticeShape = lat.shape();
      Complex currentSum = 0.0f;
      uInt nPixels = 0u;
      RO_LatticeIterator<Complex> iter(lat, 
    				   LatticeStepper(latticeShape, cursorShape));
      for (iter.reset(); !iter.atEnd(); iter++){
        currentSum += sum(iter.cursor());
        nPixels += iter.cursor().nelements();
      }
      return currentSum/nPixels;
    }

Example 2:

Sometimes it will be neccesary to access slices of a Lattice in a nearly random way. Often this can be done using the subSection commands in the LatticeStepper class. But it is also possible to use the getSlice and putSlice functions. The following example does a two-dimensional Real to Complex Fourier transform. This example is restricted to four-dimensional Arrays (unlike the previous example) and does not set up any caches (caching is currently only used with PagedArrays). So only use getSlice and putSlice when things cannot be done using LatticeIterators.

    void FFT2DReal2Complex(Lattice<Complex>& result, 
    		       const Lattice<Float>& input){
      AlwaysAssert(input.ndim() == 4, AipsError);
      const IPosition shape = input.shape();
      const uInt nx = shape(0);
      AlwaysAssert (nx > 1, AipsError);
      const uInt ny = shape(1);
      AlwaysAssert (ny > 1, AipsError);
      const uInt npol = shape(2);
      const uInt nchan = shape(3); 
      const IPosition resultShape = result.shape();
      AlwaysAssert(resultShape.nelements() == 4, AipsError);
      AlwaysAssert(resultShape(3) == nchan, AipsError);
      AlwaysAssert(resultShape(2) == npol, AipsError);
      AlwaysAssert(resultShape(1) == ny, AipsError);
      AlwaysAssert(resultShape(0) == nx/2 + 1, AipsError);
    
      const IPosition inputSliceShape(4,nx,ny,1,1);
      const IPosition resultSliceShape(4,nx/2+1,ny,1,1);
      COWPtr<Array<Float> > 
        inputArrPtr(new Array<Float>(inputSliceShape.nonDegenerate()));
      Array<Complex> resultArray(resultSliceShape.nonDegenerate());
      FFTServer<Float, Complex> FFT2D(inputSliceShape.nonDegenerate());
     
      IPosition start(4,0);
      Bool isARef;
      for (uInt c = 0; c < nchan; c++){
        for (uInt p = 0; p < npol; p++){
          isARef = input.getSlice(inputArrPtr,
                                  Slicer(start,inputSliceShape), True);
          FFT2D.fft(resultArray, *inputArrPtr);
          result.putSlice(resultArray, start);
          start(2) += 1;
        }
        start(2) = 0;
        start(3) += 1;
      }
    }

Example 3:

Occasionally you may want to access a few elements of a Lattice without all the difficulty involved in setting up Iterators or calling getSlice and putSlice. This is demonstrated in the example below and uses the parenthesis operator, along with the LatticeValueRef companion class. Using these functions to access many elements of a Lattice is not recommended as this is the slowest access method.

In this example an ideal point spread function will be inserted into an empty Lattice. As with the previous examples all the action occurs inside a function because Lattice is an interface (abstract) class.

    void makePsf(Lattice<Float>& psf) {
      const IPosition centrePos = psf.shape()/2;
      psf.set(0.0f);       // this sets all the elements to zero
                           // As it uses a LatticeIterator it is efficient
      psf(centrePos) = 1;  // This sets just the centre element to one
      AlwaysAssert(near(psf(centrePos), 1.0f, 1E-6), AipsError);
      AlwaysAssert(near(psf(centrePos*0), 0.0f, 1E-6), AipsError);
    }

Motivation

Creating an abstract base class which provides a common interface between memory and disk based arrays has a number of advantages.

It allows functions common to all arrays to be written independent of the way the data is stored. This is illustrated in the three examples above.
It reduces the learning curve for new users who only have to become familiar with one interface (ie. Lattice) rather than distinct interfaces for different array types.

Member Description

MaskedLattice() : itsDefRegPtr(0)

Default constructor.

MaskedLattice (const MaskedLattice<T>&)

Copy constructor.

virtual ~MaskedLattice()

a virtual destructor is needed so that it will use the actual destructor in the derived class

virtual MaskedLattice<T>* cloneML() const = 0
virtual Lattice<T>* clone() const

Make a copy of the object (reference semantics).

virtual Bool isMasked() const

Has the object really a mask? The default implementation returns True if the MaskedLattice has a region with a mask.

virtual Bool hasPixelMask() const

Does the lattice have a pixelmask? The default implementation returns False.

virtual const Lattice<Bool>& pixelMask() const
virtual Lattice<Bool>& pixelMask()

Get access to the pixelmask. An exception is thrown if the lattice does not have a pixelmask.

const LatticeRegion& region() const

Get the region used. This is in principle the region pointed to by getRegionPtr. However, if that pointer is 0, it returns a LatticeRegion for the full image.

Bool getMask (COWPtr<Array<Bool> >& buffer, Bool removeDegenerateAxes=False) const
Bool getMaskSlice (COWPtr<Array<Bool> >& buffer, const Slicer& section, Bool removeDegenerateAxes=False) const
Bool getMaskSlice (COWPtr<Array<Bool> >& buffer, const IPosition& start, const IPosition& shape, Bool removeDegenerateAxes=False) const
Bool getMaskSlice (COWPtr<Array<Bool> >& buffer, const IPosition& start, const IPosition& shape, const IPosition& stride, Bool removeDegenerateAxes=False) const
Bool getMask (Array<Bool>& buffer, Bool removeDegenerateAxes=False)
Bool getMaskSlice (Array<Bool>& buffer, const Slicer& section, Bool removeDegenerateAxes=False)
Bool getMaskSlice (Array<Bool>& buffer, const IPosition& start, const IPosition& shape, Bool removeDegenerateAxes=False)
Bool getMaskSlice (Array<Bool>& buffer, const IPosition& start, const IPosition& shape, const IPosition& stride, Bool removeDegenerateAxes=False)
Array<Bool> getMask (Bool removeDegenerateAxes=False) const
Array<Bool> getMaskSlice (const Slicer& section, Bool removeDegenerateAxes=False) const
Array<Bool> getMaskSlice (const IPosition& start, const IPosition& shape, Bool removeDegenerateAxes=False) const
Array<Bool> getMaskSlice (const IPosition& start, const IPosition& shape, const IPosition& stride, Bool removeDegenerateAxes=False) const

Get the mask or a slice from the mask. This is the mask formed by combination of the possible pixelmask of the lattice and the possible mask of the region taken from the lattice. If there is no mask, it still works fine. In that case it sizes the buffer correctly and sets it to True.

virtual Bool doGetMaskSlice (Array<Bool>& buffer, const Slicer& section)

The function (in the derived classes) doing the actual work. These functions are public, so they can be used internally in the various Lattice classes.
However, doGetMaskSlice does not call Slicer::inferShapeFromSource to fill in possible unspecified section values. Therefore one should normally use one of the getMask(Slice) functions. doGetMaskSlice should be used with care and only when performance is an issue.
The default implementation gets the mask from the region and fills the buffer with True values if there is no region.

MaskedLattice<T>& operator= (const MaskedLattice<T>&)

Assignment can only be used by derived classes.

virtual const LatticeRegion* getRegionPtr() const = 0

Get a pointer to the region used. It can return 0 meaning that the MaskedLattice is the full lattice.