Index of values

Absolute tolerance for the stop criterion.

anniml_predict fnet fwts finputs fpredict makes predictions on a file of input vectors finputs.

anniml_run fnet fwts input runs the network with topology file fnet and weights file fwts on an input vector input and returns an output vector y.

anniml_test fnet fwts fpatterns tests the network with topology file fnet and weights file fwts, on patterns fpatterns.

anniml_training fnet fwts fpatterns returns a couple (i,w) where i is the iteration at which the training stopped, and w is the resulting weights vector.

backprop fn nn w t (y,a_out,z,a) computes the gradient of the error function, using backpropagation of the error through the network.

bias_units nn u returns true is u is a bias unit of nn, and false otherwise.

central_diff epsilon fn nn w x t computes the gradient of the error function using central differences instead of backpropagation.

classification_results fn nn patterns w returns a tuple (sum_squares,mss,normalized_error,entropy,mce,class_mat, nb_conn,aic,mean_aic,bic,mean_bic) where sum_squares, mss, and normalized_error are respectively the raw, mean, and normalized sum of squares error, entropy is the cross entropy (see [Bishop96], section 6.9), mce is the mean cross entropy, class_mat is the matrix of correct/incorrect classifications, nb_conn is the number of unadjusted parameters (number of weights), aic is the Akaike information criterion (AIC), mean_aic is the average value of the AIC over all patterns, bic is Schwartz's Bayes information criterion (BIC), and mean_bic is the mean value of the BIC.

Columns of the patterns file that will be used to build the input vectors of the neural network.

Cross-entropy error function (see [Bishop96], section 6.9).

Derivative of the cross-entropy error function.

Derivative of the logistic function.

Derivative of the softmax function.

Derivative of the sum of squares error function.

Derivative of the hyperbolic tangent (as a function of z!).

dest nn k returns the destination unit of connection k of network nn.

Working directory.

Step of the gradient descent.

File containing input vectors only (whereas patterns files contain also target vectors).

List of float arguments found when parsing the command line.

Network file, containing the network's topology: units per layer, connections.

forward fn nn w x returns a tuple (y,a_out,z,a), where y is the output vector of the network with functions fn, connections nn, and weights w, a_out is the weighted sum of the inputs of the units belonging to the output layer, z is the output of the hidden units, and a is the weighted sum of the inputs of the hidden units.

Patterns file.

File where the network's ouptuts are saved of the -predict option is used.

fprint_list ch l prints a list l of integer on channel ch

fprint_network ch nn prints the full description of network nn on channel ch.

fprint_parameters ch prints the global variables used as program parameters on channel ch.

fprint_pattern ch (x,t) prints the input vector x and the target vector t and channel ch.

fprint_short_topology ch nn prints a short description of network nn on channel ch.

print_vector ch x prints the vector x on channel ch

Frequency of the verbose output on stdout (default 10).

fully_connected raw_layers builds a fully connected feed-forward network.

Weights file.

get_functions () returns the neural network's functions chosen by the user (command line options).

hidden layers returns option None when layers contains only one input and one output layer, and Some (u1,u2), where u1 and u2 are the first and last hidden units respectively, if there are some hidden layers.

List of integer arguments found when parsing the command line.

layer_of nn u returns the number of the layer to which unit u belongs, according to the network's descriptin in nn.

learning allows to choose how many patterns are used to compute the gradient of the error, before making a step in the descent direction (in the weights space).

Log-likelihood error function

Sigmoid exponential function logistic(x)=1/(1+exp(-x))

Maximum number of iterations.

Program's mode, which may be either Train, Test, Run, or Predict.

mu is the momentum parameter of the gradient descent with momentum method.

forward fn nn w x returns the output vector y of the network with functions fn, connections nn, and weights w.

Boolean flag deciding if the patterns should be normalized before being taken as inputs.

normalize patterns normalizes the input vectors by removing the average value and dividing by the standard deviation.

Optimization method used to minimize the error during the training.

outlayer layers returns the first and last units of the output layer.

parse_arguments () parses the arguments and the options of the command line.

Boolean flag selecting the type of results printed on the standard output: classification results when true, or regression results otherwise.

print_classification_results res prints different kinds of results (see classification_results) on the standard output.

print_regression_results res prints different kinds of results (see regression_results) on the standard output.

print_results fn nn patterns w prc_flag prints various indicators of the performance of a neural network with functions fn, topology nn, and weights w, when applied to a set of patterns patterns.

Root for the random generator (default 0).

random_weights nn returns a vector of randomly chosen weights for the network described in nn.

Number of units in each layer, without counting the bias units which are added when the fully connected topology is created.

read_inputs dim_x filename columns reads a file of input vectors of dimension dim_x, from a file filename.

read_network filename reads a neural network's description (layers, connections, and bias units) from a file, and returns the network's topology.

read_patterns dim_x dim_y filename columns reads a patterns file and returns them as an array of couples (x,t) where x is an input vector of dimension dim_x, and t is a target vector of dimension dim_y.

read_weights filename reads weights from file filename.

regression_results fn nn patterns w returns a tuple (sum_squares,mss,normalized_error,likelihood,mll,nb_conn,aic,mean_aic, bic,mean_bic) where sum_squares is the usual sum of squares error, mss is the mean sum of squares, averaged on the number of patterns, normalized_error is the sum of squares normalized by the variance of the target vectors (see section 7.5, p.

Relative tolerance for the stop criterion.

save_network nn filename saves the network's topology nn into file filename

save_weights w filename saves the vector of weights w in file filename.

Softmax function (see [Bishop96], section 6.9): softmax(a_k)= exp(a_k)/sum_j(exp(a_j).

source nn k returns the source unit of connection k of network nn.

stats patterns returns the mean value and standard deviation of the input vectors of patterns

"Sum of squares" error function.