CUV  0.9.201304091348
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Groups Pages
Enumerations | Functions
Pointwise scalar functors
Pointwise functors on vectors
Collaboration diagram for Pointwise scalar functors:

Enumerations

enum  cuv::ScalarFunctor {
  SF_EXP, SF_SIN, SF_COS, SF_LOG,
  SF_SIGN, SF_SIGM, SF_DSIGM, SF_SQUARE,
  SF_SUBLIN, SF_ENERG, SF_INV, SF_SQRT,
  SF_NEGATE, SF_ABS, SF_SMAX, SF_POSLIN,
  SF_RECT, SF_DRECT, SF_COPY, SF_LOG1P,
  SF_POW, SF_DPOW, SF_ADD, SF_SUBTRACT,
  SF_RSUB, SF_MULT, SF_DIV, SF_RDIV,
  SF_LOGADDEXP, SF_MIN, SF_MAX, SF_ROBUST_ABS,
  SF_DROBUST_ABS, SF_EQ, SF_LT, SF_GT,
  SF_LEQ, SF_GEQ, SF_BERNOULLI_KL, SF_DBERNOULLI_KL,
  SF_AXPB, SF_TANH, SF_DTANH
}
 Scalar Functors for vectors and matrices Applied pointwise to a vector/matrix. More...

Functions

template<class D >
void cuv::apply_scalar_functor (D &v, const ScalarFunctor &sf, const tensor< unsigned char, typename D::memory_space_type, typename D::memory_layout_type > *mask=NULL)
 apply a scalar functor to all elements of v.
template<class D , class S >
void cuv::apply_scalar_functor (D &dst, const S &src, const ScalarFunctor &sf, const tensor< unsigned char, typename D::memory_space_type, typename D::memory_layout_type > *mask=NULL)
 This is an overloaded member function, provided for convenience. It differs from the above function only in what argument(s) it accepts. apply a scalar functor to all elements of src.
template<class D >
void cuv::apply_scalar_functor (D &v, const ScalarFunctor &sf, const typename D::value_type &p, const tensor< unsigned char, typename D::memory_space_type, typename D::memory_layout_type > *mask=NULL)
 This is an overloaded member function, provided for convenience. It differs from the above function only in what argument(s) it accepts. in-place, one parameter.
template<class D , class S >
void cuv::apply_scalar_functor (D &dst, const S &src, const ScalarFunctor &sf, const typename S::value_type &p, const tensor< unsigned char, typename D::memory_space_type, typename D::memory_layout_type > *mask=NULL)
 This is an overloaded member function, provided for convenience. It differs from the above function only in what argument(s) it accepts.one parameter.
template<class D >
void cuv::apply_scalar_functor (D &v, const ScalarFunctor &sf, const typename D::value_type &p, const typename D::value_type &p2, const tensor< unsigned char, typename D::memory_space_type, typename D::memory_layout_type > *mask=NULL)
 This is an overloaded member function, provided for convenience. It differs from the above function only in what argument(s) it accepts.in-place, two parameters.
template<class D , class S >
void cuv::apply_scalar_functor (D &dst, const S &src, const ScalarFunctor &sf, const typename S::value_type &p, const typename S::value_type &p2, const tensor< unsigned char, typename D::memory_space_type, typename D::memory_layout_type > *mask=NULL)
 This is an overloaded member function, provided for convenience. It differs from the above function only in what argument(s) it accepts.

Detailed Description

Enumeration Type Documentation

enum cuv::ScalarFunctor

Scalar Functors for vectors and matrices Applied pointwise to a vector/matrix. 

Each entry x is transformed according to the given formular.

Without scalar parameters:

@li SF_COPY computes x (the identity)
@li SF_EXP computes exp(x)
@li SF_LOG computes log(x)
@li SF_SIGN computes sign(x)
@li SF_SIGM computes 1/(1+exp(-x))
@li SF_DSIGM computes x * (1-x)
@li SF_TANH computes tanh(x)
@li SF_DTANH computes x - x^2
@li SF_SIN computes sin(x)
@li SF_COS computes cos(x)
@li SF_SQUARE computes x*x
@li SF_SUBLIN computes 1-x
@li SF_ENERG computes -log(x) 
@li SF_INV computes 1/x 
@li SF_SQRT computes sqrt(x)
@li SF_NEGATE computes -x
@li SF_ABS computes absolute value of x
@li SF_SMAX computes (1/x -1) * x
@li SF_LOG1P computes \form#14

With one scalar parameter a:

  • SF_POW computes $x^a$
  • SF_DPOW computes $1/a * x^{a-1}$
  • SF_ADD computes x + a
  • SF_SUBTRACT computes x - a
  • SF_RSUB computes a - x
  • SF_MULT computes x * a
  • SF_DIV computes x / a
  • SF_RDIV computes a / x
  • SF_LOGADDEXP computes $\log(\exp(a) + \exp(x))$
  • SF_MIN computes min(x,a)
  • SF_MAX computes max(x,a)
  • SF_ROBUST_ABS computes $\sqrt{x^2+a}$
  • SF_DROBUST_ABS computes derivative of
    See Also
    SF_ROBUST_ABS $x/\sqrt{a+x^2}$
  • SF_EQ computes x == a
  • SF_LT computes x < a
  • SF_GT computes x > a
  • SF_LEQ computes x <= a
  • SF_GEQ computes x >= a
  • SF_BERNOULLI_KL computes Kullback-Leibler divergence of two bernoulli variables $x\log(a/x)+(1-a)\log\frac{1-a}{1-x}$
  • SF_DBERNOULLI_KL computes derivative of Kullback-Leibler divergence of two bernoulli variables w.r.t. x: $\frac{a-x}{x(x-1)}$

With two scalar parameters a and b:

  • SF_AXPB computes a*x+b

With zero OR two scalar parameters a and b:

  • SF_DTANH computes a/b * (a+x) + (a-x), if no params given: a=b=1
  • SF_TANH computes b*tanh(a*x), if no params given: a=b=1

Definition at line 192 of file tensor_ops.hpp.

Function Documentation

template<class D >
void cuv::apply_scalar_functor ( D &  v,
const ScalarFunctor &  sf,
const tensor< unsigned char, typename D::memory_space_type, typename D::memory_layout_type > *  mask = NULL 
)

apply a scalar functor to all elements of v.

Pseudocode:

forall i:
if !mask or mask[i]:
v[i] = sf(v[i]);

in-place, no parameters

Parameters
vwhere we read from and write to
sfthe applied scalar functor
maskoptional

Definition at line 290 of file tensor_ops.hpp.

template<class D , class S >
void cuv::apply_scalar_functor ( D &  dst,
const S &  src,
const ScalarFunctor &  sf,
const tensor< unsigned char, typename D::memory_space_type, typename D::memory_layout_type > *  mask = NULL 
)

This is an overloaded member function, provided for convenience. It differs from the above function only in what argument(s) it accepts. apply a scalar functor to all elements of src.

no parameters

Parameters
dstdestination
srcsource
sfthe scalar functor to be applied
maskoptional mask

Definition at line 308 of file tensor_ops.hpp.

template<class D >
void cuv::apply_scalar_functor ( D &  v,
const ScalarFunctor &  sf,
const typename D::value_type &  p,
const tensor< unsigned char, typename D::memory_space_type, typename D::memory_layout_type > *  mask = NULL 
)

This is an overloaded member function, provided for convenience. It differs from the above function only in what argument(s) it accepts. in-place, one parameter.

This function applies a binary functor, fixing the second argument to a constant.

Pseudocode:

forall i:
if !mask or mask[i]:
v[i] = bf(v[i],p);
Parameters
vsource and destination
sfthe scalar functor to be applied
pparameter for binary functor
maskoptional mask

Definition at line 333 of file tensor_ops.hpp.

template<class D , class S >
void cuv::apply_scalar_functor ( D &  dst,
const S &  src,
const ScalarFunctor &  sf,
const typename S::value_type &  p,
const tensor< unsigned char, typename D::memory_space_type, typename D::memory_layout_type > *  mask = NULL 
)

This is an overloaded member function, provided for convenience. It differs from the above function only in what argument(s) it accepts.one parameter.

This function applies a binary functor, fixing the second argument to a constant.

Pseudocode:

forall i:
if !mask or mask[i]:
dst[i] = bf(src[i],p);
Parameters
dstdestination
srcsource
sfthe scalar functor to be applied
pparameter for binary functor
maskoptional mask

Definition at line 357 of file tensor_ops.hpp.

template<class D >
void cuv::apply_scalar_functor ( D &  v,
const ScalarFunctor &  sf,
const typename D::value_type &  p,
const typename D::value_type &  p2,
const tensor< unsigned char, typename D::memory_space_type, typename D::memory_layout_type > *  mask = NULL 
)

This is an overloaded member function, provided for convenience. It differs from the above function only in what argument(s) it accepts.in-place, two parameters.

This function applies a ternary functor, fixing the second and third argument to a constant.

Pseudocode:

forall i:
if !mask or mask[i]:
v[i] = tf(v[i],p,p2);
Parameters
vsource and destination
sfthe scalar functor to be applied
p2nd parameter for ternary functor
p23rd parameter for ternary functor
maskoptional mask

Definition at line 382 of file tensor_ops.hpp.

template<class D , class S >
void cuv::apply_scalar_functor ( D &  dst,
const S &  src,
const ScalarFunctor &  sf,
const typename S::value_type &  p,
const typename S::value_type &  p2,
const tensor< unsigned char, typename D::memory_space_type, typename D::memory_layout_type > *  mask = NULL 
)

This is an overloaded member function, provided for convenience. It differs from the above function only in what argument(s) it accepts.

two parameters. This function applies a ternary functor, fixing the second and third argument to a constant.

Pseudocode:

forall i:
if !mask or mask[i]:
dst[i] = tf(src[i],p,p2);
Parameters
dstdestination
srcsource
sfthe scalar functor to be applied
p2nd parameter for ternary functor
p23rd parameter for ternary functor
maskoptional mask

Definition at line 407 of file tensor_ops.hpp.

Generated on Tue Apr 9 2013 13:48:57 for CUV by   doxygen 1.8.1.2