DFT-CASES: Generate cases from a Dirichlet diffusion tree model. Dft-cases produces a data file containing the target values for cases generated from a diffusion tree model. The diffusion tree(s) underlying these cases may optionally be displayed as well. Usage: dft-cases [ -# ] [ -b ] [ -h ] [ -p ] [ -l ] [ -t ] [ -g ] [ -n | -N ] log-file index output-file n-cases [ random-seed ] [ new-log-file ] Values for the targets in n-cases cases are generated and stored in the specified output file, one per line, overwriting any data it previously contained. If no training data has been specified, the generated cases will come from the prior distribution conditional on the overall hyperparameters and diffusion tree parameters stored in the log file under the specified index. If training data has been supplied, the cases will be generated from the posterior distribution given the training data, as represented by the hyperparameters, parameters, latent vectors, and diffusion trees(s) for training cases that are stored under that index. Including a flag of the form -#, where # is a digit from 1 up to the number of trees in the model, causes the data written to output-file to be the values generated at the terminal nodes for that tree, rather than the final values for observable variables. This allows one to examine the contributions of the different trees, since running dft-cases more than once with different values for this option, or without it, but with other arguments such as the random number seed the same, will display components of the same randomly-generated data. If one or more other flags are specified, the overall hyperparameters, diffusion tree parameters, latent vectors, tree structures, or node descriptions will be displayed on standard output, in the format documented in dft-display.doc, with the newly generated cases following the training cases (if any). By default, none of this information is displayed. When -l or -N is specified, values for latent vectors or node locations are displayed even if they are not stored in the log file - they are randomly generated if necessary (assuming that this is possible for the model being used). The random number seed used for generating the cases may be specified at the end of the command line. If it is omitted, index is used as the seed. If the final new-log-file argument is present, records of the overall hyperparameters, diffusion tree parameters, latent vectors, tree structures, and node locations for both the original cases and the newly-generated cases will be appended to the log file with this name, with an index one greater than the index of the current last record. This log file should already contain model and data specifications, which must be compatible with the records written (though this is not checked). This facility allows a new chain to be run starting from the generated state, which is useful for testing. Copyright (c) 1995-2003 by Radford M. Neal