libeblearn
|
#include <ebl_pooling.h>
Public Member Functions | |
subsampling_module (parameter< T, Tstate > *p, uint thickness, idxdim &kernel, idxdim &stride, const char *name="subsampling", bool crop=true, bool pad=true) | |
virtual | ~subsampling_module () |
destructor | |
virtual void | fprop (Tstate &in, Tstate &out) |
forward propagation from in to out | |
virtual void | bprop (Tstate &in, Tstate &out) |
backward propagation from out to in | |
virtual void | bbprop (Tstate &in, Tstate &out) |
second-derivative backward propagation from out to in | |
virtual void | forget (forget_param_linear &fp) |
forgetting weights by replacing with random values | |
virtual int | replicable_order () |
order of operation | |
virtual bool | resize_output (Tstate &in, Tstate &out) |
resize the output based on input dimensions | |
virtual fidxdim | fprop_size (fidxdim &i_size) |
virtual fidxdim | bprop_size (const fidxdim &o_size) |
virtual subsampling_module< T, Tstate > * | copy () |
Returns a deep copy of this module. | |
virtual std::string | describe () |
Returns a string describing this module and its parameters. | |
virtual void | dump_fprop (Tstate &in, Tstate &out) |
Public Attributes | |
Tstate | coeff |
Learned averaging coefficients. | |
Tstate | sub |
Temporary buffer to hold sum of neighborhood. | |
uint | thickness |
Number of features. | |
idxdim | kernel |
Dimensions of subsampling kernel. | |
idxdim | stride |
Strides of subsampling. | |
Protected Attributes | |
bool | crop |
Crop input when size mismatch or not. | |
bool | pad |
Pad output when size mismatch or not. |
This module applies 2D subsampling on dimensions 1 and 2 (0 contains different layers of information). This module has a replicable order of 3, if the input has a bigger order, use the replicable version of this module: subsampling_module_replicable.
ebl::subsampling_module< T, Tstate >::subsampling_module | ( | parameter< T, Tstate > * | p, |
uint | thickness, | ||
idxdim & | kernel, | ||
idxdim & | stride, | ||
const char * | name = "subsampling" , |
||
bool | crop = true , |
||
bool | pad = true |
||
) |
Constructor.
p | is used to store all parametric variables in a single place. If p is null, a local buffer will be used. |
thickness | The number of features. |
kernel | Size of subsampling kernel (without thickness). |
stride | Stride of subsampling kernel (without thickness). |
crop | If true, crop input when it does not match with the kernel. This allows to feed any input size to this module. |
pad | If true, add padding to end of non-feature dimensions to match the kernel when not matching. |
fidxdim ebl::subsampling_module< T, Tstate >::bprop_size | ( | const fidxdim & | o_size | ) | [virtual] |
Return dimensions compatible with this module given output dimensions. See module_1_1_gen's documentation for more details.
Just multiply each dimension by its stride
Reimplemented from ebl::module_1_1< T, Tstate >.
void ebl::subsampling_module< T, Tstate >::dump_fprop | ( | Tstate & | in, |
Tstate & | out | ||
) | [virtual] |
Calls fprop and then dumps internal buffers, inputs and outputs into files. This can be useful for debugging.
Reimplemented from ebl::module_1_1< T, Tstate >.
fidxdim ebl::subsampling_module< T, Tstate >::fprop_size | ( | fidxdim & | i_size | ) | [virtual] |
Return dimensions that are compatible with this module. See module_1_1_gen's documentation for more details.
Recompute the input size to be compliant with the output
Reimplemented from ebl::module_1_1< T, Tstate >.