[top]add_layer
In dlib, a deep neural network is composed of 3 main parts. An
input layer, a bunch of
computational layers,
and optionally a
loss layer. The add_layer
class is the central object which adds a computational layer onto an
input layer or an entire network. Therefore, deep neural networks are created
by stacking many layers on top of each other using the add_layer class.
For a tutorial showing how this is accomplished read
the DNN Introduction part 1 and
DNN Introduction part 2.
C++ Example Programs:
dnn_introduction_ex.cpp,
dnn_introduction2_ex.cpp,
dnn_inception_ex.cpp,
dnn_imagenet_ex.cpp,
dnn_imagenet_train_ex.cpp,
dnn_mmod_ex.cpp,
dnn_mmod_face_detection_ex.cpp,
dnn_mmod_dog_hipsterizer.cpp [top]add_skip_layer
This object adds a new layer to a deep neural network which draws its input
from a
tagged layer rather than from
the immediate predecessor layer as is normally done.
For a tutorial showing how to use tagging see the
dnn_introduction2_ex.cpp
example program.
[top]add_tag_layer
This object is a tool for tagging layers in a deep neural network. These tags make it
easy to refer to the tagged layer in other parts of your code.
Specifically, this object adds a new layer onto a deep neural network.
However, this layer simply performs the identity transform.
This means it is a noop and its presence does not change the
behavior of the network. It exists solely to be used by
add_skip_layer or
layer() to reference a
particular part of a network.
For a tutorial showing how to use tagging see the
dnn_introduction2_ex.cpp
example program.
C++ Example Programs:
dnn_introduction2_ex.cpp [top]alias_tensor
This object is a
tensor that
aliases another tensor. That is, it doesn't have its own block of
memory but instead simply holds pointers to the memory of another
tensor object. It therefore allows you to efficiently break a tensor
into pieces and pass those pieces into functions.
[top]approximate_distance_function
This function attempts to find a
distance_function object which is close
to a target distance_function. That is, it searches for an X such that target(X) is
minimized. Critically, X may be set to use fewer basis vectors than the target.
The optimization begins with an initial guess supplied by the user
and searches for an X which locally minimizes target(X). Since
this problem can have many local minima the quality of the starting point
can significantly influence the results.
[top]assignment_function
This object is a tool for solving the optimal assignment problem given a
user defined method for computing the quality of any particular assignment.
C++ Example Programs:
assignment_learning_ex.cpp [top]average_precision
This function computes the average precision of a ranking.
[top]batch_cached
This is a convenience function for creating
batch_trainer objects that are setup
to use a kernel matrix cache.
[top]batch_trainer
This is a batch trainer object that is meant to wrap online trainer objects
that create
decision_functions. It
turns an online learning algorithm such as
svm_pegasos
into a batch learning object. This allows you to use objects like
svm_pegasos with functions (e.g.
cross_validate_trainer)
that expect batch mode training objects.
[top]bottom_up_cluster
This function runs a bottom up agglomerative clustering algorithm.
[top]cca
This function performs a canonical correlation analysis between two sets
of vectors. Additionally, it is designed to be very fast, even for large
datasets of over a million high dimensional vectors.
[top]chinese_whispers
This function performs the clustering algorithm described in the paper
Chinese Whispers  an Efficient Graph Clustering Algorithm and its
Application to Natural Language Processing Problems by Chris Biemann.
In particular, this is a method for automatically clustering the nodes in a
graph into groups. The method is able to automatically determine the number
of clusters.
[top]compute_lda_transform
This function performs the dimensionality reducing version of linear
discriminant analysis. That is, you give it a set of labeled vectors and it
returns a linear transform that maps the input vectors into a new space that
is good for distinguishing between the different classes.
[top]compute_mean_squared_distance
This is a function that simply finds the average squared distance between all
pairs of a set of data samples. It is often convenient to use the reciprocal
of this value as the estimate of the gamma parameter of the
radial_basis_kernel.
[top]count_ranking_inversions
Given two sets of objects, X and Y, and an ordering relationship defined
between their elements, this function counts how many times we see an element
in the set Y ordered before an element in the set X. Additionally, this
routine executes efficiently in O(n*log(n)) time via the use of quick sort.
[top]cross_validate_ranking_trainer
Performs kfold cross validation on a user supplied ranking trainer object such
as the
svm_rank_trainer
and returns the fraction of ranking pairs ordered correctly as well as the mean
average precision.
C++ Example Programs:
svm_rank_ex.cppPython Example Programs:
svm_rank.py [top]cross_validate_regression_trainer
Performs kfold cross validation on a user supplied regression trainer object such
as the
svr_trainer and returns the mean squared error
and Rsquared value.
C++ Example Programs:
svr_ex.cpp [top]cross_validate_trainer_threaded
Performs kfold cross validation on a user supplied binary classification trainer object such
as the
svm_nu_trainer or
rbf_network_trainer.
This function does the same thing as
cross_validate_trainer
except this function also allows you to specify how many threads of execution to use.
So you can use this function to take advantage of a multicore system to perform
cross validation faster.
[top]decision_function
This object represents a classification or regression function that was
learned by a kernel based learning algorithm. Therefore, it is a function
object that takes a sample object and returns a scalar value.
C++ Example Programs:
svm_ex.cpp [top]discriminant_pca
This object implements the Discriminant PCA technique described in the paper:
A New Discriminant Principal Component Analysis Method with Partial Supervision (2009)
by Dan Sun and Daoqiang Zhang
This algorithm is basically a straightforward generalization of the classical PCA
technique to handle partially labeled data. It is useful if you want to learn a linear
dimensionality reduction rule using a bunch of data that is partially labeled.
[top]distance_function
This object represents a point in kernel induced feature space.
You may use this object to find the distance from the point it
represents to points in input space as well as other points
represented by distance_functions.
[top]empirical_kernel_map
This object represents a map from objects of sample_type (the kind of object
a kernel function
operates on) to finite dimensional column vectors which
represent points in the kernel feature space defined by whatever kernel
is used with this object.
To use the empirical_kernel_map you supply it with a particular kernel and a set of
basis samples. After that you can present it with new samples and it will project
them into the part of kernel feature space spanned by your basis samples.
This means the empirical_kernel_map is a tool you can use to very easily kernelize
any algorithm that operates on column vectors. All you have to do is select a
set of basis samples and then use the empirical_kernel_map to project all your
data points into the part of kernel feature space spanned by those basis samples.
Then just run your normal algorithm on the output vectors and it will be effectively
kernelized.
Regarding methods to select a set of basis samples, if you are working with only a
few thousand samples then you can just use all of them as basis samples.
Alternatively, the
linearly_independent_subset_finder
often works well for selecting a basis set. I also find that picking a
random subset typically works well.
C++ Example Programs:
empirical_kernel_map_ex.cpp,
linear_manifold_regularizer_ex.cpp [top]equal_error_rate
This function finds a threshold that best separates the elements of two
vectors by selecting the threshold with equal error rate. It also reports
the value of the equal error rate.
[top]find_clusters_using_angular_kmeans
This is a simple linear kmeans clustering implementation.
To compare a sample to a cluster, it measures the angle between them
with respect to the origin. Therefore, it tries to find clusters
of points that all have small angles between each cluster member.
[top]find_clusters_using_kmeans
This is a simple linear kmeans clustering implementation.
It uses Euclidean distance to compare samples.
[top]find_gamma_with_big_centroid_gap
This is a function that tries to pick a reasonable default value for the
gamma parameter of the
radial_basis_kernel. It
picks the parameter that gives the largest separation between the centroids, in
kernel feature space, of two classes of data.
C++ Example Programs:
rank_features_ex.cpp [top]fix_nonzero_indexing
This is a simple function that takes a std::vector of
sparse vectors
and makes sure they are zeroindexed (e.g. makes sure the first index value is zero).
[top]graph_labeler
This object is a tool for labeling each node in a
graph
with a value of true or false, subject to a labeling consistency constraint between
nodes that share an edge. In particular, this object is useful for
representing a graph labeling model learned via some machine learning
method, such as the
structural_graph_labeling_trainer.
C++ Example Programs:
graph_labeling_ex.cpp [top]histogram_intersection_kernel
This object represents a histogram intersection kernel for use with
kernel learning machines.
[top]input_rgb_image
This is a simple input layer type for use in a deep neural network
which takes an RGB image as input and loads it into a network. It
is very similar to the
input layer except that
it allows you to subtract the average color value from each color
channel when converting an image to a tensor.
[top]input_rgb_image_sized
This layer has an interface and behavior identical to
input_rgb_image
except that it requires input images to have a particular size.
[top]is_assignment_problem
This function takes a set of training data for an assignment problem
and reports back if it could possibly be a well formed assignment problem.
[top]is_binary_classification_problem
This function simply takes two vectors, the first containing feature vectors and
the second containing labels, and reports back if the two could possibly
contain data for a well formed classification problem.
[top]is_forced_assignment_problem
This function takes a set of training data for a forced assignment problem
and reports back if it could possibly be a well formed forced assignment problem.
[top]is_graph_labeling_problem
This function takes a set of training data for a graph labeling problem
and reports back if it could possibly be a well formed problem.
[top]is_learning_problem
This function simply takes two vectors, the first containing feature vectors and
the second containing labels, and reports back if the two could possibly
contain data for a well formed learning problem. In this case it just means
that the two vectors have the same length and aren't empty.
[top]is_ranking_problem
This function takes a set of training data for a learningtorank problem
and reports back if it could possibly be a well formed problem.
[top]is_sequence_labeling_problem
This function takes a set of training data for a sequence labeling problem
and reports back if it could possibly be a well formed sequence labeling problem.
[top]is_sequence_segmentation_problem
This function takes a set of training data for a sequence segmentation problem
and reports back if it could possibly be a well formed sequence segmentation problem.
[top]is_track_association_problem
This function takes a set of training data for a track association learning problem
and reports back if it could possibly be a well formed track association problem.
[top]kcentroid
This object represents a weighted sum of sample points in a kernel induced
feature space. It can be used to kernelize any algorithm that requires only
the ability to perform vector addition, subtraction, scalar multiplication,
and inner products.
An example use of this object is as an online algorithm for recursively estimating
the centroid of a sequence of training points. This object then allows you to
compute the distance between the centroid and any test points. So you can use
this object to predict how similar a test point is to the data this object has
been trained on (larger distances from the centroid indicate dissimilarity/anomalous
points).
The object internally keeps a set of "dictionary vectors"
that are used to represent the centroid. It manages these vectors using the
sparsification technique described in the paper The Kernel Recursive Least
Squares Algorithm by Yaakov Engel. This technique allows us to keep the
number of dictionary vectors down to a minimum. In fact, the object has a
user selectable tolerance parameter that controls the trade off between
accuracy and number of stored dictionary vectors.
C++ Example Programs:
kcentroid_ex.cpp [top]kernel_matrix
This is a simple set of functions that makes it easy to turn a kernel
object and a set of samples into a kernel matrix. It takes these two
things and returns a
matrix expression
that represents the kernel matrix.
[top]kkmeans
This is an implementation of a kernelized kmeans clustering algorithm.
It performs kmeans clustering by using the
kcentroid object.
If you want to use the linear kernel (i.e. do a normal kmeans clustering) then you
should use the find_clusters_using_kmeans routine.
C++ Example Programs:
kkmeans_ex.cpp [top]krls
This is an implementation of the kernel recursive least squares algorithm
described in the paper The Kernel Recursive Least Squares Algorithm by Yaakov Engel.
The long and short of this algorithm is that it is an online kernel based
regression algorithm. You give it samples (x,y) and it learns the function
f(x) == y. For a detailed description of the algorithm read the above paper.
Note that if you want to use the linear kernel then you would
be better off using the rls object as it
is optimized for this case.
C++ Example Programs:
krls_ex.cpp,
krls_filter_ex.cpp [top]learn_platt_scaling
This function is an implementation of the algorithm described in the following
papers:
Probabilistic Outputs for Support Vector Machines and Comparisons to
Regularized Likelihood Methods by John C. Platt. March 26, 1999
A Note on Platt's Probabilistic Outputs for Support Vector Machines
by HsuanTien Lin, ChihJen Lin, and Ruby C. Weng
This function is the tool used to implement the
train_probabilistic_decision_function routine.
[top]linearly_independent_subset_finder
This is an implementation of an online algorithm for recursively finding a
set (aka dictionary) of linearly independent vectors in a kernel induced
feature space. To use it you decide how large you would like the dictionary
to be and then you feed it sample points.
The implementation uses the Approximately Linearly Dependent metric described
in the paper The Kernel Recursive Least Squares Algorithm by Yaakov Engel to
decide which points are more linearly independent than others. The metric is
simply the squared distance between a test point and the subspace spanned by
the set of dictionary vectors.
Each time you present this object with a new sample point
it calculates the projection distance and if it is sufficiently large then this
new point is included into the dictionary. Note that this object can be configured
to have a maximum size. Once the max dictionary size is reached each new point
kicks out a previous point. This is done by removing the dictionary vector that
has the smallest projection distance onto the others. That is, the "least linearly
independent" vector is removed to make room for the new one.
C++ Example Programs:
empirical_kernel_map_ex.cpp [top]linear_kernel
This object represents a linear function kernel for use with
kernel learning machines.
[top]linear_manifold_regularizer
Many learning algorithms attempt to minimize a function that, at a high
level, looks like this:
f(w) == complexity + training_set_error
The idea is to find the set of parameters, w, that gives low error on
your training data but also is not "complex" according to some particular
measure of complexity. This strategy of penalizing complexity is
usually called regularization.
In the above setting, all the training data consists of labeled samples.
However, it would be nice to be able to benefit from unlabeled data.
The idea of manifold regularization is to extract useful information from
unlabeled data by first defining which data samples are "close" to each other
(perhaps by using their 3 nearest neighbors)
and then adding a term to
the above function that penalizes any decision rule which produces
different outputs on data samples which we have designated as being close.
It turns out that it is possible to transform these manifold regularized learning
problems into the normal form shown above by applying a certain kind of
preprocessing to all our data samples. Once this is done we can use a
normal learning algorithm, such as the svm_c_linear_trainer,
on just the
labeled data samples and obtain the same output as the manifold regularized
learner would have produced.
The linear_manifold_regularizer is a tool for creating this preprocessing
transformation. In particular, the transformation is linear. That is, it
is just a matrix you multiply with all your samples. For a more detailed
discussion of this topic you should consult the following paper. In
particular, see section 4.2. This object computes the inverse T matrix
described in that section.
Linear Manifold Regularization for Large Scale Semisupervised Learning
by Vikas Sindhwani, Partha Niyogi, and Mikhail Belkin
C++ Example Programs:
linear_manifold_regularizer_ex.cpp [top]load_image_dataset_metadata
dlib comes with a graphical tool for annotating images with
labeled rectangles. The tool produces an XML file containing these
annotations. Therefore, load_image_dataset_metadata() is a routine
for parsing these XML files. Note also that this is the metadata
format used by the image labeling tool included with dlib in the
tools/imglab folder.
[top]load_libsvm_formatted_data
This is a function that loads the data from a file that uses
the LIBSVM format. It loads the data into a std::vector of
sparse vectors.
If you want to load data into dense vectors (i.e.
dlib::matrix objects) then you can use the
sparse_to_dense
function to perform the conversion. Also, some LIBSVM formatted files number
their features beginning with 1 rather than 0. If this bothers you, then you
can fix it by using the
fix_nonzero_indexing function
on the data after it is loaded.
[top]lspi
This object is an implementation of the reinforcement learning algorithm
described in the following paper:
Lagoudakis, Michail G., and Ronald Parr. "Leastsquares policy
iteration." The Journal of Machine Learning Research 4 (2003):
11071149.
[top]mlp
This object represents a multilayer layer perceptron network that is
trained using the back propagation algorithm. The training algorithm also
incorporates the momentum method. That is, each round of back propagation
training also adds a fraction of the previous update. This fraction
is controlled by the momentum term set in the constructor.
It is worth noting that a MLP is, in general, very inferior to modern
kernel algorithms such as the support vector machine. So if you haven't
tried any other techniques with your data you really should.
C++ Example Programs:
mlp_ex.cppImplementations:mlp_kernel_1:
This is implemented in the obvious way.
kernel_1a  is a typedef for mlp_kernel_1 
kernel_1a_c 
is a typedef for kernel_1a that checks its preconditions.

[top]modularity
This function computes the modularity of a particular graph clustering. This
is a number that tells you how good the clustering is. In particular, it
is the measure optimized by the
newman_cluster
routine.
[top]multiclass_linear_decision_function
This object represents a multiclass classifier built out of a set of
binary classifiers. Each binary classifier is used to vote for the
correct multiclass label using a one vs. all strategy. Therefore,
if you have N classes then there will be N binary classifiers inside
this object. Additionally, this object is linear in the sense that
each of these binary classifiers is a simple linear plane.
[top]nearest_center
This function takes a list of cluster centers and a query vector
and identifies which cluster center is nearest to the query vector.
[top]newman_cluster
This function performs the clustering algorithm described in the paper
Modularity and community structure in networks by M. E. J. Newman.
In particular, this is a method for automatically clustering the nodes in a
graph into groups. The method is able to automatically determine the number
of clusters and does not have any parameters. In general, it is a very good
clustering technique.
[top]normalized_function
This object represents a container for another function
object and an instance of the
vector_normalizer object.
It automatically normalizes all inputs before passing them
off to the contained function object.
C++ Example Programs:
svm_ex.cpp [top]null_trainer_type
This object is a simple tool for turning a
decision_function
(or any object with an interface compatible with decision_function)
into a trainer object that always returns the original decision
function when you try to train with it.
dlib contains a few "training post processing" algorithms (e.g.
reduced and reduced2). These tools
take in a trainer object,
tell it to perform training, and then they take the output decision
function and do some kind of post processing to it. The null_trainer_type
object is useful because you can use it to run an already
learned decision function through the training post processing
algorithms by turning a decision function into a null_trainer_type
and then giving it to a post processor.
[top]offset_kernel
This object represents a kernel with a fixed value offset
added to it.
[top]one_vs_all_decision_function
This object represents a multiclass classifier built out
of a set of binary classifiers. Each binary classifier
is used to vote for the correct multiclass label using a
one vs. all strategy. Therefore, if you have N classes then
there will be N binary classifiers inside this object.
[top]one_vs_all_trainer
This object is a tool for turning a bunch of binary classifiers
into a multiclass classifier. It does this by training the binary
classifiers in a one vs. all fashion. That is, if you have N possible
classes then it trains N binary classifiers which are then used
to vote on the identity of a test sample.
[top]one_vs_one_decision_function
This object represents a multiclass classifier built out
of a set of binary classifiers. Each binary classifier
is used to vote for the correct multiclass label using a
one vs. one strategy. Therefore, if you have N classes then
there will be N*(N1)/2 binary classifiers inside this object.
C++ Example Programs:
multiclass_classification_ex.cpp,
custom_trainer_ex.cpp [top]one_vs_one_trainer
This object is a tool for turning a bunch of binary classifiers
into a multiclass classifier. It does this by training the binary
classifiers in a one vs. one fashion. That is, if you have N possible
classes then it trains N*(N1)/2 binary classifiers which are then used
to vote on the identity of a test sample.
C++ Example Programs:
multiclass_classification_ex.cpp,
custom_trainer_ex.cpp [top]pick_initial_centers
This is a function that you can use to seed data clustering algorithms
like the
kkmeans clustering method. What it
does is pick reasonable starting points for clustering by basically
trying to find a set of points that are all far away from each other.
C++ Example Programs:
kkmeans_ex.cpp [top]policy
This is a policy (i.e. a control law) based on a linear function approximator.
You can use a tool like
lspi to learn the parameters
of a policy.
[top]polynomial_kernel
This object represents a polynomial kernel for use with
kernel learning machines.
[top]probabilistic_decision_function
This object represents a binary decision function for use with
kernelbased learningmachines. It returns an
estimate of the probability that a given sample is in the +1 class.
C++ Example Programs:
svm_ex.cpp [top]probabilistic_function
This object represents a binary decision function for use with
any kind of binary classifier. It returns an
estimate of the probability that a given sample is in the +1 class.
[top]process_sample
This object holds a training sample for a reinforcement learning algorithm
(e.g.
lspi).
In particular, it contains a state, action, reward, next state sample from
some process.
[top]projection_function
This object represents a function that takes a data sample and projects
it into kernel feature space. The result is a real valued column vector that
represents a point in a kernel feature space. Instances of
this object are created using the
empirical_kernel_map.
C++ Example Programs:
linear_manifold_regularizer_ex.cpp [top]radial_basis_kernel
This object represents a radial basis function kernel for use with
kernel learning machines.
C++ Example Programs:
svm_ex.cpp [top]randomize_samples
Randomizes the order of samples in a column vector containing sample data.
C++ Example Programs:
svm_ex.cpp [top]ranking_pair
This object is used to contain a ranking example. Therefore, ranking_pair
objects are used to represent training examples for learningtorank tasks,
such as those used by the
svm_rank_trainer.
C++ Example Programs:
svm_rank_ex.cppPython Example Programs:
svm_rank.py [top]rank_features
Finds a ranking of the top N (a user supplied parameter) features in a set of data
from a two class classification problem. It
does this by computing the distance between the centroids of both classes in kernel defined
feature space. Good features are then ones that result in the biggest separation between
the two centroids.
C++ Example Programs:
rank_features_ex.cpp [top]rank_unlabeled_training_samples
This routine implements an active learning method for selecting the most
informative data sample to label out of a set of unlabeled samples.
In particular, it implements the MaxMin Margin and Ratio Margin methods
described in the paper:
Support Vector Machine Active Learning with Applications to Text Classification
by Simon Tong and Daphne Koller.
[top]rbf_network_trainer
Trains a radial basis function network and outputs a
decision_function.
This object can be used for either regression or binary classification problems.
It's worth pointing out that this object is essentially an unregularized version
of
kernel ridge regression. This means
you should really prefer to use kernel ridge regression instead.
[top]reduced_decision_function_trainer
This is a batch trainer object that is meant to wrap other batch trainer objects
that create
decision_function objects.
It performs post processing on the output decision_function objects
with the intent of representing the decision_function with fewer
basis vectors.
[top]reduced_decision_function_trainer2
This is a batch trainer object that is meant to wrap other batch trainer objects
that create decision_function objects.
It performs post processing on the output decision_function objects
with the intent of representing the decision_function with fewer
basis vectors.
It begins by performing the same post processing as
the reduced_decision_function_trainer
object but it also performs a global gradient based optimization
to further improve the results. The gradient based optimization is
implemented using the approximate_distance_function routine.
C++ Example Programs:
svm_ex.cpp [top]repeat
This object adds N copies of a computational layer onto a deep neural network.
It is essentially the same as using
add_layer N times,
except that it involves less typing, and for large N, will compile much faster.
C++ Example Programs:
dnn_introduction2_ex.cpp [top]resizable_tensor
This object represents a 4D array of float values, all stored contiguously
in memory. Importantly, it keeps two copies of the floats, one on the host
CPU side and another on the GPU device side. It automatically performs the
necessary host/device transfers to keep these two copies of the data in
sync.
All transfers to the device happen asynchronously with respect to the
default CUDA stream so that CUDA kernel computations can overlap with data
transfers. However, any transfers from the device to the host happen
synchronously in the default CUDA stream. Therefore, you should perform
all your CUDA kernel launches on the default stream so that transfers back
to the host do not happen before the relevant computations have completed.
If DLIB_USE_CUDA is not #defined then this object will not use CUDA at all.
Instead, it will simply store one host side memory block of floats.
Finally, the convention in dlib code is to interpret the tensor as a set of
num_samples() 3D arrays, each of dimension k() by nr() by nc(). Also,
while this class does not specify a memory layout, the convention is to
assume that indexing into an element at coordinates (sample,k,nr,nc) can be
accomplished via:
host()[((sample*t.k() + k)*t.nr() + nr)*t.nc() + nc]
[top]rls
This is an implementation of the linear version of the recursive least
squares algorithm. It accepts training points incrementally and, at
each step, maintains the solution to the following optimization problem:
find w minimizing: 0.5*dot(w,w) + C*sum_i(y_i  trans(x_i)*w)^2
Where (x_i,y_i) are training pairs. x_i is some vector and y_i is a target
scalar value.
[top]roc_c1_trainer
This is a convenience function for creating
roc_trainer_type objects that are
setup to pick a point on the ROC curve with respect to the +1 class.
[top]roc_c2_trainer
This is a convenience function for creating
roc_trainer_type objects that are
setup to pick a point on the ROC curve with respect to the 1 class.
[top]roc_trainer_type
This object is a simple trainer post processor that allows you to
easily adjust the bias term in a trained decision_function object.
That is, this object lets you pick a point on the ROC curve and
it will adjust the bias term appropriately.
So for example, suppose you wanted to set the bias term so that
the accuracy of your decision function on +1 labeled samples was 99%.
To do this you would use an instance of this object declared as follows:
roc_trainer_type<trainer_type>(your_trainer, 0.99, +1);
[top]rr_trainer
Performs linear ridge regression and outputs a decision_function that
represents the learned function. In particular, this object can only be used with
the linear_kernel. It is optimized for the linear case where
the number of features in each sample vector is small (i.e. on the order of 1000 or less since the
algorithm is cubic in the number of features.).
If you want to use a nonlinear kernel then you should use the krr_trainer.
This object is capable of automatically estimating its regularization parameter using
leaveoneout crossvalidation.
[top]rvm_regression_trainer
Trains a relevance vector machine for solving regression problems.
Outputs a decision_function that represents the learned
regression function.
The implementation of the RVM training algorithm used by this library is based
on the following paper:
Tipping, M. E. and A. C. Faul (2003). Fast marginal likelihood maximisation
for sparse Bayesian models. In C. M. Bishop and B. J. Frey (Eds.), Proceedings
of the Ninth International Workshop on Artificial Intelligence and Statistics,
Key West, FL, Jan 36.
C++ Example Programs:
rvm_regression_ex.cpp [top]rvm_trainer
Trains a relevance vector machine for solving binary classification problems.
Outputs a decision_function that represents the learned classifier.
The implementation of the RVM training algorithm used by this library is based
on the following paper:
Tipping, M. E. and A. C. Faul (2003). Fast marginal likelihood maximisation
for sparse Bayesian models. In C. M. Bishop and B. J. Frey (Eds.), Proceedings
of the Ninth International Workshop on Artificial Intelligence and Statistics,
Key West, FL, Jan 36.
C++ Example Programs:
rvm_ex.cpp [top]sammon_projection
This is a function object that computes the Sammon projection of a set
of N points in a Ldimensional vector space onto a ddimensional space
(d < L), according to the paper:
A Nonlinear Mapping for Data Structure Analysis (1969) by J.W. Sammon
[top]save_image_dataset_metadata
This routine is a tool for saving labeled image metadata to an
XML file. In particular, this routine saves the metadata into a
form which can be read by the
load_image_dataset_metadata
routine. Note also that this is the metadata
format used by the image labeling tool included with dlib in the
tools/imglab folder.
[top]save_libsvm_formatted_data
This is actually a pair of overloaded functions. Between the two of them
they let you save
sparse
or dense data vectors to file using the LIBSVM format.
[top]select_all_distinct_labels
This is a function which determines all distinct values present in a
std::vector and returns the result.
[top]sequence_labeler
This object is a tool for doing sequence labeling. In particular,
it is capable of representing sequence labeling models such as
those produced by Hidden Markov SVMs or Conditional Random fields.
See the following papers for an introduction to these techniques:
Hidden Markov Support Vector Machines by
Y. Altun, I. Tsochantaridis, T. Hofmann
Shallow Parsing with Conditional Random Fields by
Fei Sha and Fernando Pereira
C++ Example Programs:
sequence_labeler_ex.cpp [top]sequence_segmenter
This object is a tool for segmenting a sequence of objects into a set of
nonoverlapping chunks. An example sequence segmentation task is to take
English sentences and identify all the named entities. In this example,
you would be using a sequence_segmenter to find all the chunks of
contiguous words which refer to proper names.
Internally, the sequence_segmenter uses the BIO (Begin, Inside, Outside) or
BILOU (Begin, Inside, Last, Outside, Unit) sequence tagging model.
Moreover, it is implemented using a sequence_labeler
object and therefore sequence_segmenter objects are examples of
chain structured conditional random field style sequence
taggers.
C++ Example Programs:
sequence_segmenter_ex.cppPython Example Programs:
sequence_segmenter.py,
[top]shape_predictor_trainer
This object is a tool for training
shape_predictors
based on annotated training images. Its implementation uses the algorithm described in:
One Millisecond Face Alignment with an Ensemble of Regression Trees
by Vahid Kazemi and Josephine Sullivan, CVPR 2014
It is capable of learning high quality shape models. For example, this is an example output
for one of the faces in the HELEN face dataset:
C++ Example Programs:
train_shape_predictor_ex.cppPython Example Programs:
train_shape_predictor.py [top]sigmoid_kernel
This object represents a sigmoid kernel for use with
kernel learning machines.
[top]simplify_linear_decision_function
This is a set of functions that takes various forms of linear
decision functions
and collapses them down so that they only compute a single dot product when invoked.
[top]sort_basis_vectors
A kernel based learning method ultimately needs to select a set of basis functions
represented by a particular choice of kernel and a set of basis vectors.
sort_basis_vectors() is a function which attempts to perform supervised
basis set selection. In particular, you give it a candidate set of basis
vectors and it sorts them according to how useful they are for solving
a particular decision problem.
[top]sparse_histogram_intersection_kernel
This object represents a histogram intersection kernel kernel for use with
kernel learning machines that operate on
sparse vectors.
[top]sparse_linear_kernel
This object represents a linear kernel for use with
kernel learning machines that operate on
sparse vectors.
[top]sparse_polynomial_kernel
This object represents a polynomial kernel for use with
kernel learning machines that operate on
sparse vectors.
[top]sparse_radial_basis_kernel
This object represents a radial basis function kernel for use with
kernel learning machines that operate on
sparse vectors.
[top]sparse_sigmoid_kernel
This object represents a sigmoid kernel for use with
kernel learning machines that operate on
sparse vectors.
[top]spectral_cluster
This function performs the clustering algorithm described in the paper
On spectral clustering: Analysis and an algorithm by Ng, Jordan, and Weiss.
C++ Example Programs:
kkmeans_ex.cpp [top]structural_assignment_trainer
This object is a tool for learning to solve an assignment problem based
on a training dataset of example assignments. The training procedure produces an
assignment_function object which can be used
to predict the assignments of new data.
Note that this is just a convenience wrapper around the
structural_svm_assignment_problem
to make it look similar to all the other trainers in dlib.
C++ Example Programs:
assignment_learning_ex.cpp [top]structural_graph_labeling_trainer
This object is a tool for learning to solve a graph labeling problem based
on a training dataset of example labeled
graphs.
The training procedure produces a
graph_labeler object
which can be used to predict the labelings of new graphs.
To elaborate, a graph labeling problem is a task to learn a binary classifier which
predicts the label of each node in a graph. Additionally, we have information in
the form of edges between nodes where edges are present when we believe the
linked nodes are likely to have the same label. Therefore, part of a graph labeling
problem is to learn to score each edge in terms of how strongly the edge should enforce
labeling consistency between its two nodes.
Note that this is just a convenience wrapper around the
structural_svm_graph_labeling_problem
to make it look similar to all the other trainers in dlib. You might also
consider reading the book
Structured
Prediction and Learning in Computer Vision by Sebastian
Nowozin and Christoph H. Lampert since it contains a good introduction to machine learning
methods such as the algorithm implemented by the structural_graph_labeling_trainer.
C++ Example Programs:
graph_labeling_ex.cpp [top]structural_sequence_labeling_trainer
This object is a tool for learning to do sequence labeling based
on a set of training data. The training procedure produces a
sequence_labeler object which can
be use to predict the labels of new data sequences.
Note that this is just a convenience wrapper around the
structural_svm_sequence_labeling_problem
to make it look similar to all the other trainers in dlib.
C++ Example Programs:
sequence_labeler_ex.cpp [top]structural_svm_graph_labeling_problem
This object is a tool for learning the weight vectors needed to use
a
graph_labeler object.
It learns the parameter vectors by
formulating the problem as a
structural SVM problem.
[top]structural_svm_problem
This object, when used with the
oca optimizer, is a tool
for solving the optimization problem associated
with a structural support vector machine. A structural SVM is a supervised
machine learning method for learning to predict complex outputs. This is
contrasted with a binary classifier which makes only simple yes/no
predictions. A structural SVM, on the other hand, can learn to predict
complex outputs such as entire parse trees or DNA sequence alignments. To
do this, it learns a function F(x,y) which measures how well a particular
data sample x matches a label y. When used for prediction, the best label
for a new x is given by the y which maximizes F(x,y).
For an introduction to structured support vector machines you should consult
the following paper:
Predicting Structured Objects with Support Vector Machines by
Thorsten Joachims, Thomas Hofmann, Yisong Yue, and Chunnam Yu
For a more detailed discussion of the particular algorithm implemented by this
object see the following paper:
T. Joachims, T. Finley, ChunNam Yu, CuttingPlane Training of Structural SVMs,
Machine Learning, 77(1):2759, 2009.
Note that this object is essentially a tool for solving the 1Slack structural
SVM with marginrescaling. Specifically, see Algorithm 3 in the above referenced
paper.
Finally, for a very detailed introduction to this subject, you should consider the book:
Structured
Prediction and Learning in Computer Vision by Sebastian Nowozin and
Christoph H. Lampert
C++ Example Programs:
svm_struct_ex.cppPython Example Programs:
svm_struct.py,
[top]structural_svm_problem_threaded
This is just a version of the
structural_svm_problem
which is capable of using multiple cores/threads at a time. You should use it if
you have a multicore CPU and the separation oracle takes a long time to compute. Or even better, if you
have multiple computers then you can use the
svm_struct_controller_node
to distribute the work across many computers.
C++ Example Programs:
svm_struct_ex.cpp [top]structural_svm_sequence_labeling_problem
This object is a tool for learning the weight vector needed to use
a
sequence_labeler object.
It learns the parameter vector by formulating the problem as a
structural SVM problem.
The general approach is discussed in the paper:
Hidden Markov Support Vector Machines by
Y. Altun, I. Tsochantaridis, T. Hofmann
While the particular optimization strategy used is the method from:
T. Joachims, T. Finley, ChunNam Yu, CuttingPlane Training of
Structural SVMs, Machine Learning, 77(1):2759, 2009.
[top]structural_track_association_trainer
This object is a tool for learning to solve a track association problem. That
is, it takes in a set of training data and outputs a
track_association_function
you can use to do detection to track association.
C++ Example Programs:
learning_to_track_ex.cpp [top]svm_c_ekm_trainer
This object represents a tool for training the C formulation of
a support vector machine for solving binary classification problems.
It is implemented using the
empirical_kernel_map
to kernelize the
svm_c_linear_trainer. This makes it a very fast algorithm
capable of learning from very large datasets.
[top]svm_c_linear_dcd_trainer
This object represents a tool for training the C formulation of
a support vector machine to solve binary classification problems.
It is optimized for the case where linear kernels are used and
is implemented using the method described in the
following paper:
A Dual Coordinate Descent Method for Largescale Linear SVM
by ChoJui Hsieh, KaiWei Chang, and ChihJen Lin
This trainer has the ability to disable the bias term and also
to force the last element of the learned weight vector to be 1.
Additionally, it can be warmstarted from the solution to a previous
training run.
C++ Example Programs:
one_class_classifiers_ex.cpp [top]svm_c_linear_trainer
This object represents a tool for training the C formulation of
a support vector machine to solve binary classification problems.
It is optimized for the case where linear kernels are used and
is implemented using the
oca
optimizer and uses the exact line search described in the
following paper:
Optimized Cutting Plane Algorithm for LargeScale Risk Minimization
by Vojtech Franc, Soren Sonnenburg; Journal of Machine Learning
Research, 10(Oct):21572192, 2009.
This trainer has the ability to restrict the learned weights to nonnegative
values.
C++ Example Programs:
svm_sparse_ex.cpp [top]svm_c_trainer
Trains a C support vector machine for solving binary classification problems
and outputs a decision_function.
It is implemented using the SMO algorithm.
The implementation of the CSVM training algorithm used by this library is based
on the following paper:
C++ Example Programs:
svm_c_ex.cpp [top]svm_multiclass_linear_trainer
This object represents a tool for training a multiclass support
vector machine. It is optimized for the case where linear kernels
are used and implemented using the
structural_svm_problem
object.
[top]svm_nu_trainer
Trains a nu support vector machine for solving binary classification problems and
outputs a decision_function.
It is implemented using the SMO algorithm.
The implementation of the nusvm training algorithm used by this library is based
on the following excellent papers:
 Chang and Lin, Training {nu}Support Vector Classifiers: Theory and Algorithms
 ChihChung Chang and ChihJen Lin, LIBSVM : a library for support vector
machines, 2001. Software available at
http://www.csie.ntu.edu.tw/~cjlin/libsvm
C++ Example Programs:
svm_ex.cpp,
model_selection_ex.cpp [top]svm_one_class_trainer
Trains a oneclass support vector classifier and outputs a decision_function.
It is implemented using the SMO algorithm.
The implementation of the oneclass training algorithm used by this library is based
on the following paper:
C++ Example Programs:
one_class_classifiers_ex.cpp [top]svm_pegasos
This object implements an online algorithm for training a support
vector machine for solving binary classification problems.
The implementation of the Pegasos algorithm used by this object is based
on the following excellent paper:
Pegasos: Primal estimated subgradient solver for SVM (2007)
by Shai ShalevShwartz, Yoram Singer, Nathan Srebro
In ICML
This SVM training algorithm has two interesting properties. First, the
pegasos algorithm itself converges to the solution in an amount of time
unrelated to the size of the training set (in addition to being quite fast
to begin with). This makes it an appropriate algorithm for learning from
very large datasets. Second, this object uses the kcentroid object
to maintain a sparse approximation of the learned decision function.
This means that the number of support vectors in the resulting decision
function is also unrelated to the size of the dataset (in normal SVM
training algorithms, the number of support vectors grows approximately
linearly with the size of the training set).
However, if you are considering using svm_pegasos, you should also try the
svm_c_linear_trainer for linear
kernels or svm_c_ekm_trainer for nonlinear
kernels since these other trainers are, usually, faster and easier to use
than svm_pegasos.
C++ Example Programs:
svm_pegasos_ex.cpp,
svm_sparse_ex.cpp [top]svm_rank_trainer
This object represents a tool for training a ranking support vector machine
using linear kernels. In particular, this object is a tool for training
the Ranking SVM described in the paper:
Optimizing Search Engines using Clickthrough Data by Thorsten Joachims
Finally, note that the implementation of this object is done using the
oca optimizer and
count_ranking_inversions method.
This means that it runs in O(n*log(n)) time, making it suitable for use
with large datasets.
C++ Example Programs:
svm_rank_ex.cppPython Example Programs:
svm_rank.py [top]svm_struct_controller_node
This object is a tool for distributing the work involved in solving a
structural_svm_problem across many computers.
[top]svm_struct_processing_node
This object is a tool for distributing the work involved in solving a
structural_svm_problem across many computers.
[top]svr_linear_trainer
This object implements a trainer for performing epsiloninsensitive support
vector regression. It uses the
oca
optimizer so it is very efficient at solving this problem when
linear kernels are used, making it suitable for use with large
datasets.
[top]svr_trainer
This object implements a trainer for performing epsiloninsensitive support
vector regression. It is implemented using the SMO algorithm,
allowing the use of nonlinear kernels.
If you are interested in performing support vector regression with a linear kernel and you
have a lot of training data then you should use the svr_linear_trainer
which is highly optimized for this case.
The implementation of the epsSVR training algorithm used by this object is based
on the following paper:
C++ Example Programs:
svr_ex.cpp [top]test_binary_decision_function
Tests a
decision_function that represents a binary decision function and
returns the test accuracy.
[top]test_graph_labeling_function
Tests a
graph_labeler on a set of data
and returns the fraction of labels predicted correctly.
[top]test_layer
This is a function which tests if a layer object correctly implements
the
documented contract
for a computational layer in a deep neural network.
[top]test_ranking_function
Tests a
decision_function's ability to correctly
rank a dataset and returns the resulting ranking accuracy and mean average precision metrics.
C++ Example Programs:
svm_rank_ex.cppPython Example Programs:
svm_rank.py [top]test_regression_function
Tests a regression function (e.g.
decision_function)
and returns the mean squared error and Rsquared value.
[top]test_shape_predictor
Tests a
shape_predictor's ability to correctly
predict the part locations of objects. The output is the average distance (measured in pixels) between
each part and its true location. You can optionally normalize each distance using a
user supplied scale. For example, when performing face landmarking, you might want to
normalize the distances by the interocular distance.
C++ Example Programs:
train_shape_predictor_ex.cppPython Example Programs:
train_shape_predictor.py [top]track_association_function
This object is a tool that helps you implement an object tracker. So for
example, if you wanted to track people moving around in a video then this
object can help. In particular, imagine you have a tool for detecting the
positions of each person in an image. Then you can run this person
detector on the video and at each time step, i.e. at each frame, you get a
set of person detections. However, that by itself doesn't tell you how
many people there are in the video and where they are moving to and from.
To get that information you need to figure out which detections match each
other from frame to frame. This is where the track_association_function
comes in. It performs the detection to track association. It will also do
some of the track management tasks like creating a new track when a
detection doesn't match any of the existing tracks.
Internally, this object is implemented using the
assignment_function object.
In fact, it's really just a thin wrapper around assignment_function and
exists just to provide a more convenient interface to users doing detection
to track association.
C++ Example Programs:
learning_to_track_ex.cpp [top]train_probabilistic_decision_function
Trains a probabilistic_function using
some sort of binary classification trainer object such as the svm_nu_trainer or
krr_trainer.
The probability model is created by using the technique described in the following papers:
Probabilistic Outputs for Support Vector Machines and
Comparisons to Regularized Likelihood Methods by
John C. Platt. March 26, 1999
A Note on Platt's Probabilistic Outputs for Support Vector Machines
by HsuanTien Lin, ChihJen Lin, and Ruby C. Weng
C++ Example Programs:
svm_ex.cpp [top]vector_normalizer
This object represents something that can learn to normalize a set
of column vectors. In particular, normalized column vectors should
have zero mean and a variance of one.
C++ Example Programs:
svm_ex.cpp [top]vector_normalizer_frobmetric
This object is a tool for performing the FrobMetric distance metric
learning algorithm described in the following paper:
A Scalable Dual Approach to Semidefinite Metric Learning
By Chunhua Shen, Junae Kim, Lei Wang, in CVPR 2011
Therefore, this object is a tool that takes as input training triplets
(anchor, near, far) of vectors and attempts to learn a linear
transformation T such that:
length(T*anchorT*near) + 1 < length(T*anchor  T*far)
That is, you give a bunch of anchor vectors and for each anchor vector you
specify some vectors which should be near to it and some that should be far
form it. This object then tries to find a transformation matrix that makes
the "near" vectors close to their anchors while the "far" vectors are
farther away.
[top]vector_normalizer_pca
This object represents something that can learn to normalize a set
of column vectors. In particular, normalized column vectors should
have zero mean and a variance of one.
This object also uses principal component analysis for the purposes
of reducing the number of elements in a vector.
[top]verbose_batch
This is a convenience function for creating
batch_trainer objects. This function
generates a batch_trainer that will print status messages to standard
output so that you can observe the progress of a training algorithm.
C++ Example Programs:
svm_pegasos_ex.cpp [top]verbose_batch_cached
This is a convenience function for creating
batch_trainer objects. This function
generates a batch_trainer that will print status messages to standard
output so that you can observe the progress of a training algorithm.
It will also be configured to use a kernel matrix cache.