default.infact

// This is the default configuration file for the stand-alone
// executable run-model, part of the Reranker Framework (ReFr).
// It is expressed in the C++-like interpreted language that allows
// construction of primitives, objects and vectors thereof from a
// set of registered object factories.  This interpreter and associated
// code will eventually be forked as its own library called "InFact",
// hence the ".infact" filename extension.


// The path of the model file, either to read in for decoding or write out
// after training.
model_file = "example.model";

// The type of model to construct when training.
model = PerceptronModel(name("MyPerceptronModel"));

// Change this to true to use the executable in mapper mode, which specifies
// to train a single epoch and output features to standard output (mapper
// mode is used by hadoop-run.py).
mapper_mode = false;

// Uncomment this to specify training files for a training run (can
// also be done via the command line).
//training_files = { "file1", "file2" };

// Uncomment this to specify devtest files for either a training or decoding
// run (must be specified either here or on the command line, unless in
// mapper mode).
//devtest_files = { "devtest1", "devtest2" };

// Uncomment this to specify an output file when decoding (doing inference).
//output_file = "output-file";

// Uncomment this to specify an hypothesis output file when decoding (doing
// inference).
//hyp_output_file = "hyp-output-file";

// Uncomment this to do additional feature extraction on training instances.
// Put a comma-separated list of FeatureExtractor spec's in between the
// braces (may be spread across multiple lines).
//training_efe = ExecutiveFeatureExtractorImpl(extractors({ }));

// Uncomment this to do additional feature extraction on devtest instances.
// Put a comma-separated list of FeatureExtractor spec's in between the
// braces (may be spread across multiple lines).
//devtest_efe = ExecutiveFeatureExtractorImpl(extractors({ }));

// Indicates whether to re-map all feature uids to the [0,n-1] interval, where
// n is the number of non-zero features.
compactify_feature_uids = false;

// Specifies the interval after which to compactify feature uid's and
// remove unused symbols (only available when training in streaming mode).
compactify_interval = 10000;

// Specifies whether to train in streaming mode (i.e., do not read in all
// training instances into memory.
streaming = false;

// Change this to false if you are reading uncompressed input files (rare).
compressed = true;

// Change this to false if you are not use base64 encoding of input files
// (rare).
use_base64 = true;

// Change this to an integer greater than 0 to specify a minimum number of
// training epochs.
min_epochs = -1;

// Change this to an integer greater than 0 to specify a maximum number of
// training epochs.
max_epochs = -1;

// Change this to an integer greater than 0 to specify a maximum number of
// examples read from each input file.
max_examples = -1;

// Changes this to an integer greater than 0 to specify a maximum number of
// candidates to be read for each candidate set (problem instance).
max_candidates = -1;

// Specifies the interval at which the CandidateSetReader reports how
// many candidate sets it has read.
reporting_interval = 1000;

// Specifies whether to weight losses on devtest examples by the
// number of tokens in the reference, where, e.g., weighted loss is
// appropriate for computing WER, but not BLEU.
use_weighted_loss = true;