A Deep Belief Network.
A Deep Belief Network [Hinton, 2006] is a multilayer probabilistic generative models. It consists of hidden layers and visible layers. The top hidden layer and the layer below it form a Restricted Boltzmann Machine. The rest of connections in the network are directed connections that go from a hidden layer into a visible layer or another hidden layer.
The network can be pre-trained by treating it as a stack of RBMs. Each hidden layer along with the layer below it form an RBM. Each RBM is then trained using (persistent) contrastive divergence. Pre-training often provides a good initialization for the network's parameters.
After pre-training, the parameters can be fine-tuned using a variant of the wake-sleep algorithm [Hinton, 2006].
The DBN can be used to initialize the parameters of a neural network using convert_to_neural_network().
Samples can be drawn from the model by starting with a random state for the top hidden layer, performing some steps of Gibbs sampling in the top RBM to obtain the states of the top hidden layer and then using those to infer the states of the lower layers using a down-pass.
Steps for using the DBN class:
在文件 DeepBeliefNetwork.h 第 91 行定义.
Public 成员函数 | |
CDeepBeliefNetwork () | |
CDeepBeliefNetwork (int32_t num_visible_units, ERBMVisibleUnitType unit_type=RBMVUT_BINARY) | |
virtual | ~CDeepBeliefNetwork () |
virtual void | add_hidden_layer (int32_t num_units) |
virtual void | initialize_neural_network (float64_t sigma=0.01) |
virtual void | set_batch_size (int32_t batch_size) |
virtual void | pre_train (CDenseFeatures< float64_t > *features) |
virtual void | pre_train (int32_t index, CDenseFeatures< float64_t > *features) |
virtual void | train (CDenseFeatures< float64_t > *features) |
virtual CDenseFeatures < float64_t > * | transform (CDenseFeatures< float64_t > *features, int32_t i=-1) |
virtual CDenseFeatures < float64_t > * | sample (int32_t num_gibbs_steps=1, int32_t batch_size=1) |
virtual void | reset_chain () |
virtual CNeuralNetwork * | convert_to_neural_network (CNeuralLayer *output_layer=NULL, float64_t sigma=0.01) |
virtual SGMatrix< float64_t > | get_weights (int32_t index, SGVector< float64_t > p=SGVector< float64_t >()) |
virtual SGVector< float64_t > | get_biases (int32_t index, SGVector< float64_t > p=SGVector< float64_t >()) |
virtual const char * | get_name () const |
virtual CSGObject * | shallow_copy () const |
virtual CSGObject * | deep_copy () const |
virtual bool | is_generic (EPrimitiveType *generic) const |
template<class T > | |
void | set_generic () |
template<> | |
void | set_generic () |
template<> | |
void | set_generic () |
template<> | |
void | set_generic () |
template<> | |
void | set_generic () |
template<> | |
void | set_generic () |
template<> | |
void | set_generic () |
template<> | |
void | set_generic () |
template<> | |
void | set_generic () |
template<> | |
void | set_generic () |
template<> | |
void | set_generic () |
template<> | |
void | set_generic () |
template<> | |
void | set_generic () |
template<> | |
void | set_generic () |
template<> | |
void | set_generic () |
template<> | |
void | set_generic () |
void | unset_generic () |
virtual void | print_serializable (const char *prefix="") |
virtual bool | save_serializable (CSerializableFile *file, const char *prefix="") |
virtual bool | load_serializable (CSerializableFile *file, const char *prefix="") |
void | set_global_io (SGIO *io) |
SGIO * | get_global_io () |
void | set_global_parallel (Parallel *parallel) |
Parallel * | get_global_parallel () |
void | set_global_version (Version *version) |
Version * | get_global_version () |
SGStringList< char > | get_modelsel_names () |
void | print_modsel_params () |
char * | get_modsel_param_descr (const char *param_name) |
index_t | get_modsel_param_index (const char *param_name) |
void | build_gradient_parameter_dictionary (CMap< TParameter *, CSGObject * > *dict) |
virtual void | update_parameter_hash () |
virtual bool | parameter_hash_changed () |
virtual bool | equals (CSGObject *other, float64_t accuracy=0.0, bool tolerant=false) |
virtual CSGObject * | clone () |
Public 属性 | |
SGVector< int32_t > | pt_cd_num_steps |
SGVector< bool > | pt_cd_persistent |
SGVector< bool > | pt_cd_sample_visible |
SGVector< float64_t > | pt_l2_coefficient |
SGVector< float64_t > | pt_l1_coefficient |
SGVector< int32_t > | pt_monitoring_interval |
SGVector< int32_t > | pt_monitoring_method |
SGVector< int32_t > | pt_max_num_epochs |
SGVector< int32_t > | pt_gd_mini_batch_size |
SGVector< float64_t > | pt_gd_learning_rate |
SGVector< float64_t > | pt_gd_learning_rate_decay |
SGVector< float64_t > | pt_gd_momentum |
int32_t | cd_num_steps |
int32_t | monitoring_interval |
int32_t | max_num_epochs |
int32_t | gd_mini_batch_size |
float64_t | gd_learning_rate |
float64_t | gd_learning_rate_decay |
float64_t | gd_momentum |
SGIO * | io |
Parallel * | parallel |
Version * | version |
Parameter * | m_parameters |
Parameter * | m_model_selection_parameters |
Parameter * | m_gradient_parameters |
uint32_t | m_hash |
Protected 成员函数 | |
virtual void | down_step (int32_t index, SGVector< float64_t > params, SGMatrix< float64_t > input, SGMatrix< float64_t > result, bool sample_states=true) |
virtual void | up_step (int32_t index, SGVector< float64_t > params, SGMatrix< float64_t > input, SGMatrix< float64_t > result, bool sample_states=true) |
virtual void | wake_sleep (SGMatrix< float64_t > data, CRBM *top_rbm, SGMatrixList< float64_t > sleep_states, SGMatrixList< float64_t > wake_states, SGMatrixList< float64_t > psleep_states, SGMatrixList< float64_t > pwake_states, SGVector< float64_t > gen_params, SGVector< float64_t > rec_params, SGVector< float64_t > gen_gradients, SGVector< float64_t > rec_gradients) |
virtual void | load_serializable_pre () throw (ShogunException) |
virtual void | load_serializable_post () throw (ShogunException) |
virtual void | save_serializable_pre () throw (ShogunException) |
virtual void | save_serializable_post () throw (ShogunException) |
Protected 属性 | |
ERBMVisibleUnitType | m_visible_units_type |
int32_t | m_num_layers |
CDynamicArray< int32_t > * | m_layer_sizes |
SGMatrixList< float64_t > | m_states |
int32_t | m_batch_size |
SGVector< float64_t > | m_params |
int32_t | m_num_params |
SGVector< int32_t > | m_bias_index_offsets |
SGVector< int32_t > | m_weights_index_offsets |
float64_t | m_sigma |
default constructor
在文件 DeepBeliefNetwork.cpp 第 51 行定义.
CDeepBeliefNetwork | ( | int32_t | num_visible_units, |
ERBMVisibleUnitType | unit_type = RBMVUT_BINARY |
||
) |
Creates a network with one layer of visible units
num_visible_units | Number of visible units |
unit_type | Type of visible units |
在文件 DeepBeliefNetwork.cpp 第 56 行定义.
|
virtual |
在文件 DeepBeliefNetwork.cpp 第 65 行定义.
|
virtual |
Adds a layer of hidden units. The layer is connected to the layer that was added directly before it.
num_units | Number of hidden units |
在文件 DeepBeliefNetwork.cpp 第 70 行定义.
|
inherited |
Builds a dictionary of all parameters in SGObject as well of those of SGObjects that are parameters of this object. Dictionary maps parameters to the objects that own them.
dict | dictionary of parameters to be built. |
在文件 SGObject.cpp 第 597 行定义.
|
virtualinherited |
Creates a clone of the current object. This is done via recursively traversing all parameters, which corresponds to a deep copy. Calling equals on the cloned object always returns true although none of the memory of both objects overlaps.
在文件 SGObject.cpp 第 714 行定义.
|
virtual |
Converts the DBN into a neural network with the same structure and parameters. The visible layer in the DBN is converted into a CNeuralInputLayer object, and the hidden layers are converted into CNeuralLogisticLayer objects. An output layer can also be stacked on top of the last hidden layer
output_layer | An output layer |
sigma | Standard deviation of the gaussian used to initialize the parameters of the output layer |
在文件 DeepBeliefNetwork.cpp 第 360 行定义.
|
virtualinherited |
A deep copy. All the instance variables will also be copied.
在文件 SGObject.cpp 第 198 行定义.
|
protectedvirtual |
Computes the states of some layer using the states of the layer above it
在文件 DeepBeliefNetwork.cpp 第 393 行定义.
Recursively compares the current SGObject to another one. Compares all registered numerical parameters, recursion upon complex (SGObject) parameters. Does not compare pointers!
May be overwritten but please do with care! Should not be necessary in most cases.
other | object to compare with |
accuracy | accuracy to use for comparison (optional) |
tolerant | allows linient check on float equality (within accuracy) |
在文件 SGObject.cpp 第 618 行定义.
|
virtual |
Returns the bias vector of layer i
index | Layer index |
p | If specified, the bias vector is extracted from it instead of m_params |
在文件 DeepBeliefNetwork.cpp 第 558 行定义.
|
inherited |
|
inherited |
|
inherited |
|
inherited |
在文件 SGObject.cpp 第 498 行定义.
|
inherited |
Returns description of a given parameter string, if it exists. SG_ERROR otherwise
param_name | name of the parameter |
在文件 SGObject.cpp 第 522 行定义.
|
inherited |
Returns index of model selection parameter with provided index
param_name | name of model selection parameter |
在文件 SGObject.cpp 第 535 行定义.
|
virtual |
Returns the name of the SGSerializable instance. It MUST BE the CLASS NAME without the prefixed `C'.
实现了 CSGObject.
在文件 DeepBeliefNetwork.h 第 214 行定义.
|
virtual |
Returns the weights matrix between layer i and i+1
index | Layer index |
p | If specified, the weight matrix is extracted from it instead of m_params |
在文件 DeepBeliefNetwork.cpp 第 547 行定义.
|
virtual |
Initializes the DBN
sigma | Standard deviation of the gaussian used to initialize the weights |
在文件 DeepBeliefNetwork.cpp 第 76 行定义.
|
virtualinherited |
If the SGSerializable is a class template then TRUE will be returned and GENERIC is set to the type of the generic.
generic | set to the type of the generic if returning TRUE |
在文件 SGObject.cpp 第 296 行定义.
|
virtualinherited |
Load this object from file. If it will fail (returning FALSE) then this object will contain inconsistent data and should not be used!
file | where to load from |
prefix | prefix for members |
在文件 SGObject.cpp 第 369 行定义.
|
protectedvirtualinherited |
Can (optionally) be overridden to post-initialize some member variables which are not PARAMETER::ADD'ed. Make sure that at first the overridden method BASE_CLASS::LOAD_SERIALIZABLE_POST is called.
ShogunException | will be thrown if an error occurs. |
被 CKernel, CWeightedDegreePositionStringKernel, CList, CAlphabet, CLinearHMM, CGaussianKernel, CInverseMultiQuadricKernel, CCircularKernel , 以及 CExponentialKernel 重载.
在文件 SGObject.cpp 第 426 行定义.
|
protectedvirtualinherited |
Can (optionally) be overridden to pre-initialize some member variables which are not PARAMETER::ADD'ed. Make sure that at first the overridden method BASE_CLASS::LOAD_SERIALIZABLE_PRE is called.
ShogunException | will be thrown if an error occurs. |
被 CDynamicArray< T >, CDynamicArray< float64_t >, CDynamicArray< float32_t >, CDynamicArray< int32_t >, CDynamicArray< char >, CDynamicArray< bool > , 以及 CDynamicObjectArray 重载.
在文件 SGObject.cpp 第 421 行定义.
|
virtualinherited |
在文件 SGObject.cpp 第 262 行定义.
|
virtual |
Pre-trains the DBN as a stack of RBMs
features | Input features. Should have as many features as the number of visible units in the DBN |
在文件 DeepBeliefNetwork.cpp 第 149 行定义.
|
virtual |
Pre-trains a single RBM
index | Index of the RBM |
features | Input features. Should have as many features as the total number of visible units in the DBN |
在文件 DeepBeliefNetwork.cpp 第 159 行定义.
|
inherited |
prints all parameter registered for model selection and their type
在文件 SGObject.cpp 第 474 行定义.
|
virtualinherited |
|
virtual |
Resets the state of the markov chain used for sampling
在文件 DeepBeliefNetwork.cpp 第 352 行定义.
|
virtual |
Draws samples from the marginal distribution of the visible units. The sampling starts from the values DBN's internal state and result of the sampling is stored there too.
num_gibbs_steps | Number of Gibbs sampling steps for the top RBM. |
batch_size | Number of samples to be drawn. A seperate chain is used for each sample |
在文件 DeepBeliefNetwork.cpp 第 333 行定义.
|
virtualinherited |
Save this object to file.
file | where to save the object; will be closed during returning if PREFIX is an empty string. |
prefix | prefix for members |
在文件 SGObject.cpp 第 314 行定义.
|
protectedvirtualinherited |
Can (optionally) be overridden to post-initialize some member variables which are not PARAMETER::ADD'ed. Make sure that at first the overridden method BASE_CLASS::SAVE_SERIALIZABLE_POST is called.
ShogunException | will be thrown if an error occurs. |
被 CKernel 重载.
在文件 SGObject.cpp 第 436 行定义.
|
protectedvirtualinherited |
Can (optionally) be overridden to pre-initialize some member variables which are not PARAMETER::ADD'ed. Make sure that at first the overridden method BASE_CLASS::SAVE_SERIALIZABLE_PRE is called.
ShogunException | will be thrown if an error occurs. |
被 CKernel, CDynamicArray< T >, CDynamicArray< float64_t >, CDynamicArray< float32_t >, CDynamicArray< int32_t >, CDynamicArray< char >, CDynamicArray< bool > , 以及 CDynamicObjectArray 重载.
在文件 SGObject.cpp 第 431 行定义.
|
virtual |
Sets the number of train/test cases the RBM will deal with
batch_size | Batch size |
在文件 DeepBeliefNetwork.cpp 第 135 行定义.
|
inherited |
在文件 SGObject.cpp 第 41 行定义.
|
inherited |
在文件 SGObject.cpp 第 46 行定义.
|
inherited |
在文件 SGObject.cpp 第 51 行定义.
|
inherited |
在文件 SGObject.cpp 第 56 行定义.
|
inherited |
在文件 SGObject.cpp 第 61 行定义.
|
inherited |
在文件 SGObject.cpp 第 66 行定义.
|
inherited |
在文件 SGObject.cpp 第 71 行定义.
|
inherited |
在文件 SGObject.cpp 第 76 行定义.
|
inherited |
在文件 SGObject.cpp 第 81 行定义.
|
inherited |
在文件 SGObject.cpp 第 86 行定义.
|
inherited |
在文件 SGObject.cpp 第 91 行定义.
|
inherited |
在文件 SGObject.cpp 第 96 行定义.
|
inherited |
在文件 SGObject.cpp 第 101 行定义.
|
inherited |
在文件 SGObject.cpp 第 106 行定义.
|
inherited |
在文件 SGObject.cpp 第 111 行定义.
|
inherited |
set generic type to T
|
inherited |
|
inherited |
|
inherited |
|
virtualinherited |
A shallow copy. All the SGObject instance variables will be simply assigned and SG_REF-ed.
被 CGaussianKernel 重载.
在文件 SGObject.cpp 第 192 行定义.
|
virtual |
Trains the DBN using the variant of the wake-sleep algorithm described in [A Fast Learning Algorithm for Deep Belief Nets, Hinton, 2006].
features | Input features. Should have as many features as the total number of visible units in the DBN |
在文件 DeepBeliefNetwork.cpp 第 210 行定义.
|
virtual |
Applies the DBN as a features transformation
Forward-propagates the input features through the DBN and returns the Mean activations of the \( i^{th} \) hidden layer
features | Input features. Should have as many features as the number of visible units in the DBN |
i | Index of the hidden layer. If -1, the activations of the last hidden layer is returned |
在文件 DeepBeliefNetwork.cpp 第 316 行定义.
|
inherited |
unset generic type
this has to be called in classes specializing a template class
在文件 SGObject.cpp 第 303 行定义.
|
protectedvirtual |
Computes the states of some layer using the states of the layer below it
在文件 DeepBeliefNetwork.cpp 第 443 行定义.
|
virtualinherited |
Updates the hash of current parameter combination
在文件 SGObject.cpp 第 248 行定义.
|
protectedvirtual |
Computes the gradients using the wake-sleep algorithm
在文件 DeepBeliefNetwork.cpp 第 473 行定义.
int32_t cd_num_steps |
Number of Gibbs sampling steps performed before each weight update during wake-sleep training. Default value is 1.
在文件 DeepBeliefNetwork.h 第 306 行定义.
float64_t gd_learning_rate |
Gradient descent learning rate for wake-sleep training, defualt value 0.1
在文件 DeepBeliefNetwork.h 第 325 行定义.
float64_t gd_learning_rate_decay |
Gradient descent learning rate decay for wake-sleep training. The learning rate is updated at each iteration i according to: alpha(i)=decay*alpha(i-1) Default value is 1.0 (no decay)
在文件 DeepBeliefNetwork.h 第 332 行定义.
int32_t gd_mini_batch_size |
Size of the mini-batch used during gradient descent wake-sleep training, If 0 full-batch training is performed Default value is 0
在文件 DeepBeliefNetwork.h 第 322 行定义.
float64_t gd_momentum |
gradient descent momentum multiplier for wake-sleep training
default value is 0.9
For more details on momentum, see this paper [Sutskever, 2013]
在文件 DeepBeliefNetwork.h 第 342 行定义.
|
inherited |
io
在文件 SGObject.h 第 369 行定义.
|
protected |
Number of train/test cases the network is currently dealing with
在文件 DeepBeliefNetwork.h 第 358 行定义.
|
protected |
Index at which the bias of each layer is stored in the parameters vector
在文件 DeepBeliefNetwork.h 第 367 行定义.
|
inherited |
parameters wrt which we can compute gradients
在文件 SGObject.h 第 384 行定义.
|
inherited |
Hash of parameter values
在文件 SGObject.h 第 387 行定义.
|
protected |
Size of each layer
在文件 DeepBeliefNetwork.h 第 352 行定义.
|
inherited |
model selection parameters
在文件 SGObject.h 第 381 行定义.
|
protected |
Number of layers
在文件 DeepBeliefNetwork.h 第 349 行定义.
|
protected |
Number of parameters
在文件 DeepBeliefNetwork.h 第 364 行定义.
|
inherited |
parameters
在文件 SGObject.h 第 378 行定义.
Parameters of the network
在文件 DeepBeliefNetwork.h 第 361 行定义.
|
protected |
Standard deviation of the gaussian used to initialize the parameters
在文件 DeepBeliefNetwork.h 第 376 行定义.
|
protected |
States of each layer
在文件 DeepBeliefNetwork.h 第 355 行定义.
|
protected |
Type of the visible units
在文件 DeepBeliefNetwork.h 第 346 行定义.
|
protected |
Index at which the weights of each hidden layer is stored in the parameters vector
在文件 DeepBeliefNetwork.h 第 372 行定义.
int32_t max_num_epochs |
Maximum number of iterations over the training set during wake-sleep training. Defualt value is 1
在文件 DeepBeliefNetwork.h 第 316 行定义.
int32_t monitoring_interval |
Number of weight updates between each evaluation of the reconstruction error during wake-sleep training. Default value is 10.
在文件 DeepBeliefNetwork.h 第 311 行定义.
|
inherited |
parallel
在文件 SGObject.h 第 372 行定义.
SGVector<int32_t> pt_cd_num_steps |
CRBM::cd_num_steps for pre-training each RBM. Default value is 1 for all RBMs
在文件 DeepBeliefNetwork.h 第 246 行定义.
SGVector<bool> pt_cd_persistent |
CRBM::cd_persistent for pre-training each RBM. Default value is true for all RBMs
在文件 DeepBeliefNetwork.h 第 251 行定义.
SGVector<bool> pt_cd_sample_visible |
CRBM::cd_sample_visible for pre-training each RBM. Default value is false for all RBMs
在文件 DeepBeliefNetwork.h 第 256 行定义.
CRBM::gd_learning_rate for pre-training each RBM. Default value is 0.1 for all RBMs
在文件 DeepBeliefNetwork.h 第 291 行定义.
CRBM::gd_learning_rate_decay for pre-training each RBM. Default value is 1.0 for all RBMs
在文件 DeepBeliefNetwork.h 第 296 行定义.
SGVector<int32_t> pt_gd_mini_batch_size |
CRBM::gd_mini_batch_size for pre-training each RBM. Default value is 0 for all RBMs
在文件 DeepBeliefNetwork.h 第 286 行定义.
CRBM::gd_momentum for pre-training each RBM. Default value is 0.9 for all RBMs
在文件 DeepBeliefNetwork.h 第 301 行定义.
CRBM::l1_coefficient for pre-training each RBM. Default value is 0.0 for all RBMs
在文件 DeepBeliefNetwork.h 第 266 行定义.
CRBM::l2_coefficient for pre-training each RBM. Default value is 0.0 for all RBMs
在文件 DeepBeliefNetwork.h 第 261 行定义.
SGVector<int32_t> pt_max_num_epochs |
CRBM::max_num_epochs for pre-training each RBM. Default value is 1 for all RBMs
在文件 DeepBeliefNetwork.h 第 281 行定义.
SGVector<int32_t> pt_monitoring_interval |
CRBM::monitoring_interval for pre-training each RBM. Default value is 10 for all RBMs
在文件 DeepBeliefNetwork.h 第 271 行定义.
SGVector<int32_t> pt_monitoring_method |
CRBM::monitoring_method for pre-training each RBM. Default value is RBMMM_RECONSTRUCTION_ERROR for all RBMs
在文件 DeepBeliefNetwork.h 第 276 行定义.
|
inherited |
version
在文件 SGObject.h 第 375 行定义.