Neighborhood components analysis (nca)
nca [-h] [-v] -i string -o string [-A double] [-l string] [-L] [-n int] [-T int] [-M double] [-m double] [-N] [-B int] [-O string] [-s int] [-a double] [-t double] [-V] [-w double]
This program implements Neighborhood Components Analysis, both a linear dimensionality reduction technique and a distance learning technique. The method seeks to improve k-nearest-neighbor classification on a dataset by scaling the dimensions. The method is nonparametric, and does not require a value of k. It works by using stochastic ("soft") neighbor assignments and using optimization techniques over the gradient of the accuracy of the neighbor assignments.
To work, this algorithm needs labeled data. It can be given as the last row of the input dataset (--input_file), or alternatively in a separate file (--labels_file).
This implementation of NCA uses either stochastic gradient descent or the L_BFGS optimizer. Both of these optimizers do not guarantee global convergence for a nonconvex objective function (NCA's objective function is nonconvex), so the final results could depend on the random seed or other optimizer parameters.
Stochastic gradient descent, specified by --optimizer "sgd", depends primarily on two parameters: the step size (--step_size) and the maximum number of iterations (--max_iterations). In addition, a normalized starting point can be used (--normalize), which is necessary if many warnings of the form 'Denominator of p_i is 0!' are given. Tuning the step size can be a tedious affair. In general, the step size is too large if the objective is not mostly uniformly decreasing, or if zero-valued denominator warnings are being issued. The step size is too small if the objective is changing very slowly. Setting the termination condition can be done easily once a good step size parameter is found; either increase the maximum iterations to a large number and allow SGD to find a minimum, or set the maximum iterations to 0 (allowing infinite iterations) and set the tolerance (--tolerance) to define the maximum allowed difference between objectives for SGD to terminate. Be careful -- setting the tolerance instead of the maximum iterations can take a very long time and may actually never converge due to the properties of the SGD optimizer.
The L-BFGS optimizer, specified by --optimizer "lbfgs", uses a back-tracking line search algorithm to minimize a function. The following parameters are used by L-BFGS: --num_basis (specifies the number of memory points used by L-BFGS), --max_iterations, --armijo_constant, --wolfe, --tolerance (the optimization is terminated when the gradient norm is below this value), --max_line_search_trials, --min_step and --max_step (which both refer to the line search routine). For more details on the L-BFGS optimizer, consult either the MLPACK L-BFGS documentation (in lbfgs.hpp) or the vast set of published literature on L-BFGS.
By default, the SGD optimizer is used.
--input_file (-i) [string] Input dataset to run NCA on.
--output_file (-o) [string] Output file for learned distance matrix.
--armijo_constant (-A) [double] Armijo constant for L-BFGS. Default value 0.0001.
--help (-h) Default help info.
--info [string] Get help on a specific module or option. Default value ''.
--labels_file (-l) [string] File of labels for input dataset. Default value ''.
--linear_scan (-L) Don't shuffle the order in which data points are visited for SGD.
--max_iterations (-n) [int] Maximum number of iterations for SGD or L-BFGS (0 indicates no limit). Default value 500000.
--max_line_search_trials (-T) [int] Maximum number of line search trials for L-BFGS. Default value 50.
--max_step (-M) [double] Maximum step of line search for L-BFGS. Default value 1e+20.
--min_step (-m) [double] Minimum step of line search for L-BFGS. Default value 1e-20.
--normalize (-N) Use a normalized starting point for optimization. This is useful for when points are far apart, or when SGD is returning NaN.
--num_basis (-B) [int] Number of memory points to be stored for L-BFGS. Default value 5.
--optimizer (-O) [string] Optimizer to use; "sgd" or "lbfgs". Default value 'sgd'.
--seed (-s) [int] Random seed. If 0, 'std::time(NULL)' is used. Default value 0.
--step_size (-a) [double] Step size for stochastic gradient descent (alpha). Default value 0.01.
--tolerance (-t) [double] Maximum tolerance for termination of SGD or L-BFGS. Default value 1e-07.
--verbose (-v) Display informational messages and the full list of parameters and timers at the end of execution.
--version (-V) Display the version of mlpack.
--wolfe (-w) [double] Wolfe condition parameter for L-BFGS. Default value 0.9.
For further information, including relevant papers, citations, and theory, consult the documentation found at http://www.mlpack.org or included with your distribution of MLPACK.