The Complete SPRLIB & ANNLIB

Index

A

• ActEucDist
  - activation function
• ActInprod
  - activation function
• ActSqEucDist
  - activation function
• adapt_mlnet
  - perform one learning cycle on a maximum likelihood network
• adapt_wavnet
  - adapt a wavelet network after presentation of one learning sample
• add_classerror
  - add a pattern to the error estimation routines using Fisher's linear discriminant
• add_mean_var
  - add a pattern to the summations needed for mean and variance
• add_unit_ff
  - add a new unit to a layer
• addq_classerror
  - add a pattern to the error estimation routines using a quadratic discriminant
• all_scatter_dataset
  - determine the scatter matrices for a dataset

B

• best_matching_unit
  - find the unit with the smallest output value in a Kohonen net
• BP-flags
  - back-propagation option flags
• bp_adapt_net
  - perform one learning cycle using the backpropagation rule
• bp_adapt_unit
  - adapt a unit according to the generalized delta rule
• bp_adapt_weights
  - adapt weights marked ``changed'' according to the accumulated delta's
• bp_free
  - free a network that was generated by bp_init
• bp_init
  - initialise and set up a network for training with the backpropagation learning rule
• bp_learn
  - perform a number of learning cycles with the backpropagation learning algorithm
• bp_learn_pocket
  - train a network using the pocket algorithm
• brent
  - parabolic interpolation and brent's method in one dimension

C

• calc_classerror
  - calculate the fraction of badly classified samples (linear classifier)
• Calc_Dist_Class
  - calculate the minimal Mahalanobis distance of a pattern given the coefficients
• calc_mean_var
  - calculate the mean and variance of patterns
• calc_multiclass_error
  - calculate the error in classification using the Mahalanobis distance
• calcq_classerror
  - calculate the fraction of badly classified samples (quadratic classifier)
• calloc_monitor
  - an SPRANNLIB replacement for the calloc routine
• CDES-flags
  - option flags for conjugate gradient descent learning
• check_all_network
  - check the consistency of a network
• clear_auto_priority
  - de-install a signal handler which can renice the process
• clear_user_signal
  - reset the original SIGUSR1 signal handler
• cluster_pao
  - perfom Pao clustering
• cluster_pao_init
  - initialize the variables in order to perform Pao clustering
• compute_means
  - compute the means of the classes in the dataset
• create_ff_net
  - create a feedforward network
• create_koh_net
  - create a Kohonen type network
• create_mlnet
  - create a maximum likelihood network
• create_sharednet
  - create a shared weights neural network
• create_wavnet
  - create a wavelet network

D

• DATASET-ACCESS-flags
  - write enable/disable system/user-data in a dataset/sample
• DATASET-flags
  - data set type specification flags
• DATASET
  - structure holding general data set information
• delete_dataset
  - free a complete dataset and all its samples
• delete_net
  - delete and free memory of a complete network
• delete_theta_hist
  - delete a network's bias (theta) history
• delete_unit_hist
  - delete a network's unit history
• delete_weight_hist
  - delete a network's weight history
• DerDerTransDblSigmoid
  - second derivative of transfer function
• DerDerTransParTanH
  - second derivative of transfer function
• DerDerTransSigmoid
  - second derivative of transfer function
• DerTransDblSigmoid
  - first derivative of transfer function
• DerTransGauss
  - first derivative of transfer function
• DerTransLinear
  - first derivative of transfer function
• DerTransParTanH
  - first derivative of transfer function
• DerTransSigmoid
  - first derivative of transfer function
• DerTransTanH
  - first derivative of transfer function
• dfpmin
  - perform Broyden-Fletcher-Goldfarb-Shanno variant of Davidon-Fletcher-Powell minimazation
• DISTRIBUTION
  - data structure holding a data set's distribution information
• duin_dataset
  - generate samples according to the Duin dataset
• duinset_error
  - compute the classification error on the Duin set using a linear discriminant
• duinset_overlap
  - compute the overlap of two classes for the Duin set

E

• eigsrt
  - sort the given eigenvectors
• erfcc
  - compute the complementary error function erfc(x) with fractional error everywhere less than 0.00000012
• ERROR_STRUCT
  - network performance values datatype
• ESTERROR-variables
  - local variables used in the error estimation routines.
• ESTIMATE-variables
  - local variables used in the error estimation routines.
• ESTQERROR-variables
  - local variables used in the error estimation routines.
• euclidian_distance
  - calculate the euclidian distance between two arrays of doubles
• eval_error_ptron
  - compute the error of a perceptron on the allocated dataset
• eval_ff_net
  - evaluate a feed-forward network
• eval_koh_net
  - evaluate a Kohonen typ network
• eval_ptron
  - evaluate a perceptron
• eval_set_on_net
  - determine the similarity between a (trained) SOM and a dataset
• eval_unit
  - evaluate a unit
• EXIT-flags
  - flags used in the SPRANNLIB exiting function
• extract_dataset_ff_net
  - extract a dataset from a certain layer of anetwork

F

• linmin
• FILE-IO-variables
  - variables used in some general file I/O functions
• find_full_fname
  - search for a file
• fisher
  - compute the optimal linear discriminant function for two classes
• fisher_dataset
  - compute the fisher discriminant function using a dataset
• fisherq
  - determine the coefficients for the quadratic decision function
• fisherq_dataset
  - compute the decision function parameters on a given dataset
• fpe_handler
  - floating point exception handler
• fprintf_compressed_dataset
  - write a compressed dataset to a file
• fprintf_compressed_network
  - save a compressed network
• fprintf_compressed_network_small
  - writes a compressed network in small I/O-format
• fprintf_dataset
  - write a complete dataset to a stream
• fprintf_error_struct
  - prints an ERROR_STRUCT 's contents
• fprintf_matrix
  - print a matrix to the file pointed by stream
• fprintf_network
  - save a network
• fprintf_network_small
  - writes a network in small I/O-format
• fprintf_ptron
  - print the perceptron data to a stream
• fprintf_vector
  - print a vector to the file pointed by stream
• free_dataset
  - deallocate a DATASET structure
• free_layer
  - frees a LAYER structure allocated using malloc_layer
• free_link
  - frees a LINK structure allocated using malloc_link
• free_map
  - frees a MAP structure allocated using malloc_map
• free_monitor
  - an SPRANNLIB replacement for the free routine
• free_net
  - frees a NET structure allocated using malloc_net
• free_ptron
  - free the allocated space
• free_sample
  - deallocate a SAMPLE structure
• free_sharednet
  - frees a shared weights neural network
• free_sharednet_recipe
  - free are SHAREDNET_RECIPE structure
• free_system_dataset
  - free all allocated SAMPLE and DATASET structures
• free_system_netlib
  - frees all memory allocated with malloc_netlib
• free_unit
  - frees a UNIT structure allocated using malloc_unit
• free_unit_value
  - frees a UNIT_VALUE structure allocated using malloc_unit_value
• free_wavnet
  - free a wavelet network
• free_weight
  - frees a WEIGHT structure allocated using malloc_weight
• free_weight_value
  - frees a WEIGHT_VALUE structure allocated using malloc_weight_value
• frprmn
  - perform Fletcher-Reeves-Polak-Ribiere minimazation
• fscanf_compressed_dataset
  - read a compressed dataset from a file
• fscanf_compressed_network
  - load a compressed network
• fscanf_compressed_network_small
  - reads a compressed network in small I/O-format
• fscanf_dataset
  - read a complete dataset from a stream
• fscanf_ioversion_net
  - get the network I/O format version number from a file
• fscanf_network
  - load a network
• fscanf_network_small
  - reads a network in small I/O-format

G

• GASMULT-variables
  - local variable used in the Gaussian random pattern functions
• get_act_func_name
  - convert an activation function pointer to a name
• get_act_func_ptr
  - convert an activation function name to a pointer
• get_inlink
  - maps two unit IDs to a link pointer
• get_map
  - maps a map ID to a pointer
• get_outlink
  - maps two unit IDs to a link pointer
• get_trans_func_name
  - convert a transfer function pointer to a name
• get_trans_func_no_pars
  - find the number of parameters of a transfer function
• get_trans_func_ptr
  - convert a transfer function name to a pointer
• get_unit
  - maps a unit ID to a pointer
• get_weight
  - maps two unit IDs to a weight pointer
• get_weight_byid
  - maps a weight ID to a weight pointer
• GetLabels
  - get the labels of a dataset
• GETOPT-variables
  - some variables used in the getopt function
• getopt
  - an MS-DOS parameter parsing routine
• grad_net_mse
  - calculate the gradient of the MSE of a dataset with respect to all parameters of a network

H

• highley_error
  - compute the error of a classifier on the Higleyman classes
• HIST-flags
  - general history flags

I

• INIT-flags
  - flags used in the SPRANNLIB initialization function
• init_full_fname
  - initializes the filename expansion mechanism
• init_mlnet
  - initialize a maximum likelihood network
• init_wavnet
  -initialize a wavelet network
• IO-variables
  - variable supporting network I/O functions
• IOFORMAT
  - constant indicating the network I/O-format version number
• ioset_to_labelset
  - convert an IOSET to a LABELSET
• isodata_basic
  - the isodata clustering algorithm as defined by Duda and Hart (k-means clustering)

J

• jacobi
  - compute the eigenvalues and eigenvectors of a real symmetric matrix
• join_net
  - construct a new network out of a number of networks

K

• khighley_dataset
  - generate samples according to the generalized Highleyman dataset
• khighley_overlap
  - compute the overlap of two classes for the Higleyman set
• kl_coordinates
  - compute the Karhunen-Loeve transform
• knn_classify
  - classify a new pattern using the k-nearest neighbour classifier
• knn_free
  - freeing routine for the k-nearest neighbour classifier
• knn_init
  - initialization routine for the k-nearest neighbour classifier
• knn_loo_perf
  - k-nearest neighbour classifier performance determination using the leave one out method
• knn_majority
  - assign a classlabel according to the k-nearest neighbour rule
• KNNELM
  - local structure for the k-nearest neighbour classifier
• koh_adapt_net
  - adapt a net's weights according to the Kohonen learning rule
• koh_adapt_unit
  - adapt a unit's weights according to the Kohonen learning rule
• koh_draw_free
  - clear all data used in drawing Kohonen maps
• koh_draw_init
  - initialize the variables for drawing a Kohonen map
• koh_learn
  - perform a number of learning cycles with the Kohonen learning rule
• koh_test_net
  - constructs a two-dimensional Kohonen network with two inputs
• koh_weight_draw
  - draw a Kohonen map
• KOHONEN-DRAW-flags
  - option flags for Kohonen weight drawing
• KOHONEN-DRAW-variables
  - variables used for graphical display of Kohonen type networks
• KOHONEN-flags
  - option flags for Kohonen learning
• kvar_dataset
  - generate samples according to the generalized var dataset
• kvar_overlap
  - compute the overlap of two classes for the varset

L

• labelset_to_ioset
  - convert a LABELSET to an IOSET
• LAYER-flags
  - flags to indicate the type of a layer
• LAYER
  - layer datatype
• LAYER_RECIPE
  - simplified description of a layer of a shared weights network
• learn_bfgs_ffnet
  - apply the BFGS optimization method to train a feedforward neural network
• learn_cgdes_ffnet
  - apply the conjugate gradient descent method to a feedforward neural network
• learn_marquardt_ffnet
  - train a feedforward or radial-basis function network with the dataset pointed by dset for steps number of times, using the Marquardt optimization method
• learn_mlnet
  - perform a number of learning cycles on a maximum likelihood network
• learn_ptron
  - train a perceptron with the loaded training samples
• learn_ptron_free
  - free the space allocated for perceptron learning
• learn_ptron_init
  - initialization routine for perceptron training
• learn_wavnet
  - train a wavelet network on a dataset
• LIBRARY-flags
  - various constants influencing library operation
• LINK-flags
  - flags to indicate the type of a link
• LINK
  - link (unit connection) datatype
• linmin
  - line minimization
• load_dataset
  - read a (possibly) compressed dataset from a file
• load_network
  - load a (compressed) network
• log_lhood_mlnet
  - determine the log-likelihood function of all samples in a dataset
• loo_min_dist_perf
  - compute the performance of the minimum distance using the leave-one-out method
• lubksb
  - solve a set of n linear equations
• ludcmp
  - lu decomposition of a matrix

M

• mahal_multiclass_dataset
  - determine the Mahalanobis coefficients for all classes in the set
• make_matrix3D
  - construct a transformation matrix to display three-dimensional data
• make_source_ffnet
  - write a stand-alone C program containing a network evaluation routine
• make_spcost_matrix
  - compute the shortest-path distance matrix between all nodes in a SOM
• malloc_dataset
  - allocate a DATASET structure
• malloc_layer
  - allocate a LAYER structure from a pool
• malloc_link
  - allocate a LINK structure from a pool
• malloc_map
  - allocate a MAP structure from a pool
• malloc_monitor
  - an SPRANNLIB replacement for the malloc routine
• malloc_net
  - allocate a NET structure from a pool
• malloc_ptron
  - allocate space for a perceptron and initialize
• malloc_sample
  - allocate a SAMPLE structure
• malloc_unit
  - allocate a UNIT structure from a pool
• malloc_unit_value
  - allocate a UNIT_VALUE structure from a pool
• malloc_weight
  - allocate a WEIGHT structure from a pool
• malloc_weight_value
  - allocate a WEIGHT_VALUE structure from a pool
• MAP
  - unit group datatype
• MAP_RECIPE
  - simplified description of a map of a shared weights network
• matcopy
  - copy matrix A to B
• matinv
  - invert a matrix A into matrix B
• matinvd
  - invert a matrix A into matrix B and return the determinant
• matmult
  - multiply two matrices and store result in a new matrix
• matrix
  - a Numerical Recipes in C datatype
• mattransp
  - transpose matrix A to At
• matvecmult
  - multiply a matrix with a vector and store the result in a new vector
• mean_var_dataset
  - estimate the mean and variance of a class in a data set
• meanset_error
  - compute the classification error on the meanset using a linear discriminant
• memory_monitor
  - return the total amount of dynamically allocated memory
• MESSAGE-functions
  - front-ends to the sprmessage function
• ML_STRUCT
  - a local structure for maximum likelihood learning
• MNBRAK
  - search for a bracketed minimum of a function in downhill direction
• MONITOR-ALLOCATION-functions
  - sprlib memory allocation routines replacing the original C-functions

N

• nearest_neighbour
  - compute the class label of the nearest neighbour
• NET-ACCESS-flags
  - flags used to regulate access to system data and userdata fields in networks
• NET-flags
  - flags to indicate the type of a net
• NET
  - network datatype
• net_mse
  - calculate the MSE of a dataset with respect to a network
• net_perf
  - compute the performance of a network on a dataset
• nn_performance
  - compute the leave-one-out performance of the nearest neighbour classifier
• norm_weights_ff
  - normalize the weights on the input links of one or more layers
• normalize_dataset
  - normalize the input and output vector of each sample
• normalize_input_dataset
  - normalize the input vector of each sample
• normalize_one_input_dataset
  - transform one element of the input vector of each sample
• normalize_one_output_dataset
  - transform one element of the output vector of each sample
• normalize_output_dataset
  - normalize the output vector of each sample

P

• PAO-variables
  - local variables used for Pao clustering
• PARZELM
  - local structure for Parzen classification
• parzen_best_s
  - determine the best smoothing value for a class
• parzen_classify
  - two class Parzen classifier
• parzen_dataset_class
  - classify a sample, given a dataset, using the parzen estimator
• parzen_dset_best_s
  - compute the optimal smoothing parameter for a class
• parzen_init
• parzen_init
  - initialization for the Parzen classifier
• parzen_mclass
  - pattern classification using the Parzen classifier
• parzen_probability
  - determine the probability of a new pattern using the Parzen classifier
• PERF-flags
  - network performance calculation option flags
• perf_free
  - free the performance testing facilities of a learning set
• perf_init
  - initialize the performance testing facilities of a learning set
• perf_mlnet
  - auxiliary function to determine the performance of a maximum likelihood network
• pn_adapt_net
  - perform one learning cycle with the pseudo-Newton variation of backpropagation
• pn_adapt_unit
  - adapt a unit according to the pseudo-Newton variation on the generalized delta rule
• pn_adapt_weights
  - adapt weights marked ``changed'' according to the accumulated (pseudo-Newton) delta's
• pn_free
  - free a network that was generated by pn_init
• pn_init
  - initialise and set up a network for training with the pseudo-Newton backpropagation learning rule
• pn_learn
  - perform a number of learning cycles with the pseudo-Newton learning algorithm
• pr_cum_dbl_exp
  - compute the cumulative (integrated) probability function
• pr_cum_normal
  - compute the cumulative probability distribution function of a standardized normal variable
• pr_dbl_exp
  - the probability density function for the double exponential distribution
• pr_error
  - compute the classification error for a one-dimensional classifier
• pr_error_multi
  - compute the classification error for a classifier
• pr_k_normal
  - compute the probability density for a general k-dimensional normal distribution
• pr_normal
  - compute the probability density of a standardized normal random variable
• PTRON-flags
  - flags for perceptron learning
• PTRON
  - a structure for perceptron learning

R

• ran1
  - "minimal" random number generator of Park and Miller with Bays-Durham shuffle
• random_dbl_exp
  - determine a random number using the double exponential function
• random_duinset_a
  - generate a sample of class A using the Duin set
• random_duinset_b
  - generate a sample of class B using the Duin set
• random_highley_a
  - generate a sample of class A according to the Highleyman distribution
• random_highley_b
  - generate a sample of class B according to the Highleyman distribution
• random_highley_dataset
  - generates a dataset which is distributed accorded to Highleyman
• random_khighley_a
  - generate a sample from class A using the khighley dataset
• random_khighley_b
  - generate a sample from class B using the khighley dataset
• random_kvar_a
  - generate a sample from class A using the kvar dataset
• random_kvar_b
  - generate a sample from class B using the kvar dataset
• random_meanset_a
  - generate a sample of class A using the mean set
• random_meanset_b
  - generate a sample of class B using the mean set
• random_multi_init
• random_multi_gauss
  - generate a random pattern using a Gauss distribution
• random_multi_init
  - initialization routine for generating a random pattern
• random_order_dataset
  - randomize the order of the samples in a dataset
• random_seed
  - initialize the random generator with a seed
• random_unif
  - front-end to ran1
• random_varset_a
  - generate a sample from class A using the var dataset
• random_varset_b
  - generate a sample from class B using the var dataset
• read_sharednet_recipe
  - read a recipe for creating a shared weights neural network
• realloc_monitor
  - an SPRANNLIB replacement for the realloc routine
• restore_pocket_weights
  - put the weights in a pocket into the network
• rss_min_dist_perf
  - compute the performance of the nearest mean classifier using the resubstitution method

S

• SAMANN-flags
  - SAMANN option flags
• samann_adapt_net
  - perform one learning cycle using the SAMANN rule
• samann_adapt_unit
  - adapt a unit according to SAMANN learning rule
• samann_free
  - free a network that was generated by samann_init
• samann_init
  - initialise and set up a network for training with the SAMANN backpropagation learning rule
• samann_learn
  - perform a number of learning cycles with the backpropagation learning algorithm
• samann_learn_sample
  - perform one learning cycles with the SAMANN learning algorithm
• samann_scale_dataset
  - scale a dataset to make it suitable for SAMANN backpropagation training
• SAMMON-flags
  - flags for Sammon map routines
• SAMMON-MAP-flags
  - flags for Sammon map injection/extraction routines
• sammon_constant
  - query the Sammon constant
• sammon_dstress
  - calculate the derivative of Sammon's stress measure
• sammon_exit
  - free memory allocated by sammon_init
• sammon_extract_map
  - extract the current Sammon mapping in the form ofa DATASET
• sammon_init
  - initialize the Sammon mapping routines
• sammon_inject_map
  - replace the current Sammon mapping by a DATASET
• sammon_rand
  - initialize a Sammon map with random values
• sammon_solve
  - minimize Sammon's stress measure
• sammon_stress
  - calculate Sammon's stress measure
• sammon_triangulate_set
  - map a dataset using an existing Sammon mapping
• SAMPLE-flags
  - enabling or disabling a data sample
• SAMPLE
  - structure holding information of specific data point
• sample_distance
  - compute the Euclidean between two SPRANNLIB samples
• sample_mse
  - calculate the MSE of one sample and the corresponding network output
• sample_perf
  - compute the performance of a network on a sample
• scale_output_dataset
  - scale one output element of an IOSET dataset
• scawi_ffnet
  - initialize a feedforward neural net using the SCAWI method
• set_auto_priority
  - install a signal handler which can renice the process
• set_user_signal
  - set the user definable signal SIGUSR1 to a userdefined function
• SHAREDNET_RECIPE
  - simplified description of a shared weights network
• som_goodness
  - compute the goodness-of-fit between a trained SOM and a dataset
• SOUND-flags
  - some flags for playing sound samples
• SOUND-variables
  - some variables used in the sound sample routines
• sound_sample
  - play a sound sample to the audio device
• sprexit
  - terminate the use of SPRANNLIB
• sprinit
  - initialize SPRANNLIB and print version and copyright information
• sprmessage
  - a general stdout error message / warning function
• STATISTICS
  - data structure with all statistics about data set
• status_layer
  - print a LAYER structure's contents to a stream
• status_link
  - print a LINK structure's contents to a stream
• status_map
  - print a MAP structure's contents to a stream
• status_monitor
  - print a memory allocation status report
• status_net
  - print a NET structure's contents to a stream
• status_total
  - print a network's contents to a stream
• status_unit
  - print a UNIT structure's contents to a stream
• status_weight
  - print a WEIGHT structure's contents to a stream
• std_mean_dataset
  - generate samples according to the mean dataset
• std_var_dataset
  - generate samples according to the var dataset
• store_hist_all_thetas
  - store the current values of the thetas in the linked lists of their history
• store_hist_all_units
  - store the current unit values in the linked list of their history
• store_hist_all_weights
  - store the current weight values in the linked lists of their history
• store_hist_theta
  - store the current value of theta in a newly allocated structure in the linked list of values
• store_hist_unit
  - store the current state of a unit in a newly allocated structure in the linked list of unit states
• store_hist_weight
  - store the current weight value in a newly allocated structure in the linked list of weight values
• subspace_dataset
  - create a new dataset from another dataset with a higher dimension
• svdcmp
  - perform a single value decomposition of matrix
• system_data
  - a union used to store external data
• system_time
  - get information about used CPU time

T

• time_stamp
  - store current time and date into a character string
• TimerHandler
  - a signal handler which can renice the process
• train_som
  - a generic front-end for SOM-training
• TransDblSigmoid
  - transfer function
• transform_output_dataset
  - transform one output element of an IOSET dataset
• TransGauss
  - transfer function
• TransLinear
  - transfer function
• TransParTanH
  - transfer function
• TransSigmoid
  - transfer function
• TransTanH
  - transfer function

U

• unif_fanin_rand_net
  - randomize a network's weights with uniformly distributed samples
• unif_jogg_net
  - randomly jogg a network's weights with uniformly distributed samples
• unif_rand_net
  - randomize a network's weights with uniformly distributed samples
• UNIT-flags
  - flags to indicate the type of a unit
• UNIT
  - unit datatype
• UNIT_VALUE
  - unit value datatype

V

• valid_activation_function
  - check if an activation function pointer is valid
• valid_der_der_transfer_function
  - check if a second derivative transfer function pointer is valid
• valid_der_transfer_function
  - check if a derivative transfer function pointer is valid
• valid_trans_func
  - check if a transfer function pointer is valid
• valid_transfer_function
  - check if a transfer function pointer is valid
• VARIOUS-constants
  - some global constants
• VARIOUS-macros
  - some global macro's
• varset_error
  - calculate the classification error for a linear discriminant function
• vecinmult
  - calculate the innner product ('dot product') of two vectors
• vector
  - a Numerical Recipes in C datatype

W

• warning_system_load
  - get system load averages and warn the user if necessary
• WEIGHT-flags
  - flags to indicate the type of a weight
• WEIGHT
  - connection weight datatype
• WEIGHT_VALUE
  - connection weight value datatype
• weight_vector
  - compute the Fisher discriminant vector
• wien_output_ff
  - calculate the Wiener solution for an output unit