PRMAPPING

Mapping class constructor

   W = PRMAPPING(MAPPING_FILE,MAPPING_TYPE,DATA,LABELS,SIZE_IN,SIZE_OUT)
   U = PRMAPPING(MAPPING_FILE,'untrained',PARS)
   W = PRMAPPING(MAPPING_FILE,'combiner',PARS)

Description

A mapping/classifier object W is constructed. It points to the command  MAPPING_FILE that is able to execute the mapping. It may be used to map a  dataset A on another dataset B by B = PRMAP(A,W) or by training a mapping  using an untrained mapping U and a dataset A: W = PRMAP(A,U) or by modifying,  (or combining) a mapping W with another mapping V: W_NEW = PRMAP(V,W); These operations may also be written as B = A*W, W = A*U or W_NEW = V*W.

As a general rule for the case A is a cell array, B = A*W is executed  for all elements in A, returning a cell array B: B{i} = A{i}*W. An  exeception is the mapping type 'fixed_cell', see below.

As PRTools contains many predifined mappings there is no need for the  starting user to define his own mappings. Advanced users may inspect  simple examples like SIGM or SUBSC to see how they are constructed.

MAPPING_FILE	name of the routine used for defining, training or  executing the mapping. Such a routine (e.g.  'mapfile'), typically a classifier or a data mapping,  should generate a mapping W or U, and should also be  able to accept and execute the following types of calls,  generated by PRTools depending on the value of  MAPPING_TYPE: MAPPING_TYPE = 'untrained': An untrained mapping U is trained by  W = mapfile(A,U) in which U is typically defined by  U = mapfile([],PARS{:}) or by just U = mapfile. PRTools  generates W = mapfile(A,U) if the user supplies  V = A*U. MAPPING_TYPE = 'trained': A trained mapping W can be applied to a  dataset D by D = mapfile(B,W), resulting in a  dataset D. PRTools generates this call if the user  supplies D = B*W. If W is the result of training an  untrained mapping U by a dataset A it holds that  D = B*(A*U). MAPPING_TYPE = 'combiner': A combiner mapping W is able to modify or  combine a set of mappings V. PRTools calls V2 = V1*W as V2 = mapfile(V1,PARS{:}). If W would not have been  a combiner then V2 = V1*W would have been stored as a  sequential mapping. See SEQUENTIAL MAPPING_TYPE = 'fixed': D = A*W is executed by D = mapfile(A,PARS{:}).  In practice there is not much of a difference between  a trained and a fixed mapping. The first is found from  data, the latter is defined directly by its parameters MAPPING_TYPE = 'fixed_cell': Like a fixed mapping. D = A*W is executed  by D = mapfile(A,PARS{:}). However, in this case A can  be a cell array. It thereby overrides the rule that a  cellarray applied to a mapping is executed element by  element.
MAPPING_TYPE	String defining the type of mapping:  'untrained', 'trained', "combiner' or 'fixed', see above.  Default is 'untrained'. PRMAPPING(MAPPING_FILE,DATA) is  equivalent to PRMAPPING(MAPPING_FILE,'untrained',DATA)
DATA	Data, structure or cell array necessary for defining the  mapping, e.g. the weights of a neural network. DATA is  just used in the MAPPING_FILE for executing the mapping.  For fixed and untrained mappings DATA can only be a  cell array.
LABELS	Array with labels to be used as feature labels for the  dataset resulting by executing the mapping. So at least  as many labels as defined by SIZE_OUT has to be supplied.
SIZE_IN	Input dimensionality or size vector describing its shape,  e.g. in case the input space is derived from an image.  For a classifier SIZE_IN is the feature size.
SIZE_OUT	Output dimensionality or size vector describing its  shape, e.g. in case the output space should represent an  image. For a classifier SIZE_OUT is the number of  classes. Default is the number of labels in LABELS.  SIZE_IN and SIZE_OUT are just used for error checking.  If SIZE_IN is not supplied they are both set to 0 and  checking is skipped.

Other parameter fields may be set to define the mapping further by

   W = PRMAPPING(MAPPING_FILE, MAPPING_TYPE, DATA, LABELS, ...
      'field1',V1,'field2',V2, ...)
  or by

   W = PRMAPPING(MAPPING_FILE, MAPPING_TYPE, DATA, LABELS, ...
      SIZE_IN, SIZE_OUT,'field1',V1,'field2',V2, ...)

The following fields are possible (if not set defaults are supplied)

SCALE	Output multiplication factor. If SCALE is a scalar all  multiplied by it. SCALE may also be a vector with size  as defined by SIZE_OUT to set separate scalings for each  output.
OUT_CONV	0,1,2,3 for defining the desired output conversion:  0 - no(default), 1: SIGM, 2: NORMM or 3: SIGM and NORMM.  These values are set by cnormc in case of 2-class  discriminants (OUTCONV = 1) and by CLASSC (OUT_CONV = OUT_CONV+2) to convert densities and  sigmoidal outputs to normalised posterior probabilities.
COST	Classification costs in case the mapping defines a  classifier. See SETCOST.
NAME	String with mapping name
USER	User definable variable

All parameters are stored in fields corresponding to the above names.  Parameter fields of a given mapping may also be changed by

   W = SET(W,'field1',V1,'field2',V2, ...)

They may also be set by the routines SETMAPPING_FILE, SETMAPPING_TYPE,  SETDATA, SETLABELS, SETSIZE_IN, SETSIZE_OUT, SETSIZE, SETSCALE, SETOUT_CONV,  SETCOST, SETNAME and SETUSER. Fields may be retrieved by

   VARARGOUT = GET(W,'field1','field2', ...)

or by the routines GETMAPPING_FILE, GETMAPPING_TYPE, GETDATA, GETSIZE_IN,  GETSIZE_OUT, GETSCALE, GETOUTCONV, GETCOST, GETNAME and GETUSER.

See also

datasets, mappings, sequential,