ConvNet
1.0
A GPU-based C++ implementation of Convolutional Neural Nets
|
Stochastic gradient descent. More...
#include <optimizer.h>
Public Member Functions | |
SGDOptimizer (const config::Optimizer &optimizer_config) | |
virtual void | AllocateMemory (const int rows, const int cols) |
virtual void | Optimize (Matrix &gradient, Matrix ¶meter) |
virtual void | LoadParameters (hid_t file, const string &prefix) |
virtual void | SaveParameters (hid_t file, const string &prefix) |
virtual bool | IsAllocated () |
Public Member Functions inherited from Optimizer | |
Optimizer (const config::Optimizer &optimizer_config) | |
virtual void | ReduceLearningRate (float factor) |
Protected Member Functions | |
float | GetMomentum () const |
Protected Member Functions inherited from Optimizer | |
float | GetDecayedEpsilon () const |
void | ApplyConstraints (Matrix ¶meter) |
Protected Attributes | |
Matrix | gradient_history_ |
const float | gradient_clip_ |
const float | initial_momentum_ |
const float | final_momentum_ |
const int | momentum_transition_timescale_ |
Protected Attributes inherited from Optimizer | |
const config::Optimizer::Decay | epsilon_decay_type_ |
float | epsilon_ |
float | minimum_epsilon_ |
const int | epsilon_decay_timescale_ |
const int | start_optimization_after_ |
const float | l2_decay_ |
const float | weight_norm_limit_ |
const float | weight_norm_constraint_ |
int | step_ |
Additional Inherited Members | |
Static Public Member Functions inherited from Optimizer | |
static Optimizer * | ChooseOptimizer (const config::Optimizer &config) |
Stochastic gradient descent.
Implements stochastic gradient descent with momentum.