From 93e2729cbc6909378694f93d18c9badb018f527e Mon Sep 17 00:00:00 2001 From: Chris Dyer Date: Thu, 20 Feb 2014 18:20:41 -0500 Subject: new defaults in cxx file --- training/mira/kbest_cut_mira.cc | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'training') diff --git a/training/mira/kbest_cut_mira.cc b/training/mira/kbest_cut_mira.cc index e075bed3..1a6415be 100644 --- a/training/mira/kbest_cut_mira.cc +++ b/training/mira/kbest_cut_mira.cc @@ -82,14 +82,14 @@ bool InitCommandLine(int argc, char** argv, po::variables_map* conf) { ("optimizer,o",po::value()->default_value(1), "Optimizer (SGD=1, PA MIRA w/Delta=2, Cutting Plane MIRA=3, PA MIRA=4, Triple nbest list MIRA=5)") ("fear,f",po::value()->default_value(1), "Fear selection (model-cost=1, maxcost=2, maxscore=3)") ("hope,h",po::value()->default_value(1), "Hope selection (model+cost=1, mincost=2)") - ("max_step_size,C", po::value()->default_value(0.01), "regularization strength (C)") + ("max_step_size,C", po::value()->default_value(0.001), "regularization strength (C)") ("random_seed,S", po::value(), "Random seed (if not specified, /dev/random will be used)") ("mt_metric_scale,s", po::value()->default_value(1.0), "Amount to scale MT loss function by") ("sent_approx,a", "Use smoothed sentence-level BLEU score for approximate scoring") ("pseudo_doc,e", "Use pseudo-document BLEU score for approximate scoring") ("no_reweight,d","Do not reweight forest for cutting plane") ("no_select,n", "Do not use selection heuristic") - ("k_best_size,k", po::value()->default_value(250), "Size of hypothesis list to search for oracles") + ("k_best_size,k", po::value()->default_value(500), "Size of hypothesis list to search for oracles") ("update_k_best,b", po::value()->default_value(1), "Size of good, bad lists to perform update with") ("unique_k_best,u", "Unique k-best translation list") ("stream,t", "Stream mode (used for realtime)") -- cgit v1.2.3 From 5935df4adc4d0b8864ae0f9a65b8f6453a11bb45 Mon Sep 17 00:00:00 2001 From: Michael Denkowski Date: Wed, 26 Feb 2014 12:15:28 -0800 Subject: Use same number of jobs for decoding. --- training/mira/mira.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'training') diff --git a/training/mira/mira.py b/training/mira/mira.py index 77f2f35f..ca549ed8 100755 --- a/training/mira/mira.py +++ b/training/mira/mira.py @@ -242,7 +242,7 @@ def evaluate(testset, weights, ini, script_dir, out_dir): evaluator = '{}/../utils/decode-and-evaluate.pl'.format(script_dir) try: p = subprocess.Popen([evaluator, '-c', ini, '-w', weights, '-i', testset, - '-d', out_dir], stdout=subprocess.PIPE) + '-d', out_dir, '--jobs', args.jobs], stdout=subprocess.PIPE) results, err = p.communicate() bleu, results = results.split('\n',1) except subprocess.CalledProcessError: -- cgit v1.2.3