summaryrefslogtreecommitdiff
path: root/training
diff options
context:
space:
mode:
Diffstat (limited to 'training')
-rw-r--r--training/dtrain/dtrain.cc4
-rw-r--r--training/dtrain/dtrain.h6
-rw-r--r--training/dtrain/examples/standard/cdec.ini2
-rw-r--r--training/dtrain/examples/standard/expected-output115
-rw-r--r--training/dtrain/examples/toy/cdec.ini1
-rw-r--r--training/dtrain/examples/toy/dtrain.ini2
-rwxr-xr-xtraining/dtrain/parallelize.rb19
-rw-r--r--training/mira/kbest_cut_mira.cc8
-rwxr-xr-xtraining/mira/mira.py19
-rwxr-xr-xtraining/pro/pro.pl24
10 files changed, 113 insertions, 87 deletions
diff --git a/training/dtrain/dtrain.cc b/training/dtrain/dtrain.cc
index b01cf421..ccb50af2 100644
--- a/training/dtrain/dtrain.cc
+++ b/training/dtrain/dtrain.cc
@@ -438,7 +438,7 @@ main(int argc, char** argv)
score_t model_diff = it->first.model - it->second.model;
score_t loss = max(0.0, -1.0 * model_diff);
- if (check && ki == 1) cout << losses[pair_idx] - loss << endl;
+ if (check && ki==repeat-1) cout << losses[pair_idx] - loss << endl;
pair_idx++;
if (repeat > 1) {
@@ -455,7 +455,7 @@ main(int argc, char** argv)
margin = fabs(model_diff);
if (!rank_error && margin < loss_margin) margin_violations++;
}
- if (rank_error && ki==1) rank_errors++;
+ if (rank_error && ki==0) rank_errors++;
if (scale_bleu_diff) eta = it->first.score - it->second.score;
if (rank_error || margin < loss_margin) {
SparseVector<weight_t> diff_vec = it->first.f - it->second.f;
diff --git a/training/dtrain/dtrain.h b/training/dtrain/dtrain.h
index eb23b813..07bd9b65 100644
--- a/training/dtrain/dtrain.h
+++ b/training/dtrain/dtrain.h
@@ -116,11 +116,11 @@ inline ostream& _p(ostream& out) { return out << setiosflags(ios::showpos); }
inline ostream& _p2(ostream& out) { return out << setprecision(2); }
inline ostream& _p5(ostream& out) { return out << setprecision(5); }
-inline void printWordIDVec(vector<WordID>& v)
+inline void printWordIDVec(vector<WordID>& v, ostream& os=cerr)
{
for (unsigned i = 0; i < v.size(); i++) {
- cerr << TD::Convert(v[i]);
- if (i < v.size()-1) cerr << " ";
+ os << TD::Convert(v[i]);
+ if (i < v.size()-1) os << " ";
}
}
diff --git a/training/dtrain/examples/standard/cdec.ini b/training/dtrain/examples/standard/cdec.ini
index 6cba9e1e..3330dd71 100644
--- a/training/dtrain/examples/standard/cdec.ini
+++ b/training/dtrain/examples/standard/cdec.ini
@@ -21,7 +21,7 @@ feature_function=RuleIdentityFeatures
feature_function=RuleSourceBigramFeatures
feature_function=RuleTargetBigramFeatures
feature_function=RuleShape
-feature_function=RuleWordAlignmentFeatures
+feature_function=LexicalFeatures 1 1 1
#feature_function=SourceSpanSizeFeatures
#feature_function=SourceWordPenalty
#feature_function=SpanFeatures
diff --git a/training/dtrain/examples/standard/expected-output b/training/dtrain/examples/standard/expected-output
index fa831221..2460cfbb 100644
--- a/training/dtrain/examples/standard/expected-output
+++ b/training/dtrain/examples/standard/expected-output
@@ -4,7 +4,8 @@ Reading ./nc-wmt11.en.srilm.gz
----5---10---15---20---25---30---35---40---45---50---55---60---65---70---75---80---85---90---95--100
****************************************************************************************************
Example feature: Shape_S00000_T00000
-Seeding random number sequence to 4138446869
+T=1 I=1 D=1
+Seeding random number sequence to 2327685089
dtrain
Parameters:
@@ -36,87 +37,87 @@ Iteration #1 of 3.
. 10
Stopping after 10 input sentences.
WEIGHTS
- Glue = -80.3
- WordPenalty = -51.247
- LanguageModel = +282.46
- LanguageModel_OOV = -85.8
- PhraseModel_0 = -100.06
- PhraseModel_1 = -98.692
- PhraseModel_2 = -9.4958
- PhraseModel_3 = +18.535
- PhraseModel_4 = +62.35
- PhraseModel_5 = +7
- PhraseModel_6 = +31.4
- PassThrough = -126.5
+ Glue = +6.9
+ WordPenalty = -46.426
+ LanguageModel = +535.12
+ LanguageModel_OOV = -123.5
+ PhraseModel_0 = -160.73
+ PhraseModel_1 = -350.13
+ PhraseModel_2 = -187.81
+ PhraseModel_3 = +172.04
+ PhraseModel_4 = +0.90108
+ PhraseModel_5 = +21.6
+ PhraseModel_6 = +67.2
+ PassThrough = -149.7
---
- 1best avg score: 0.25631 (+0.25631)
- 1best avg model score: -4843.6 (-4843.6)
- avg # pairs: 744.4
+ 1best avg score: 0.23327 (+0.23327)
+ 1best avg model score: -9084.9 (-9084.9)
+ avg # pairs: 780.7
avg # rank err: 0 (meaningless)
avg # margin viol: 0
k-best loss imp: 100%
- non0 feature count: 1274
+ non0 feature count: 1389
avg list sz: 91.3
- avg f count: 143.72
-(time 0.4 min, 2.4 s/S)
+ avg f count: 146.2
+(time 0.37 min, 2.2 s/S)
Iteration #2 of 3.
. 10
WEIGHTS
- Glue = -117.4
- WordPenalty = -99.584
- LanguageModel = +395.05
- LanguageModel_OOV = -136.8
- PhraseModel_0 = +40.614
- PhraseModel_1 = -123.29
- PhraseModel_2 = -152
- PhraseModel_3 = -161.13
- PhraseModel_4 = -76.379
- PhraseModel_5 = +39.1
- PhraseModel_6 = +137.7
- PassThrough = -162.1
+ Glue = -43
+ WordPenalty = -22.019
+ LanguageModel = +591.53
+ LanguageModel_OOV = -252.1
+ PhraseModel_0 = -120.21
+ PhraseModel_1 = -43.589
+ PhraseModel_2 = +73.53
+ PhraseModel_3 = +113.7
+ PhraseModel_4 = -223.81
+ PhraseModel_5 = +64
+ PhraseModel_6 = +54.8
+ PassThrough = -331.1
---
- 1best avg score: 0.26751 (+0.011198)
- 1best avg model score: -10061 (-5216.9)
- avg # pairs: 639.1
+ 1best avg score: 0.29568 (+0.062413)
+ 1best avg model score: -15879 (-6794.1)
+ avg # pairs: 566.1
avg # rank err: 0 (meaningless)
avg # margin viol: 0
k-best loss imp: 100%
- non0 feature count: 1845
+ non0 feature count: 1931
avg list sz: 91.3
- avg f count: 139.88
-(time 0.35 min, 2.1 s/S)
+ avg f count: 139.89
+(time 0.33 min, 2 s/S)
Iteration #3 of 3.
. 10
WEIGHTS
- Glue = -101.1
- WordPenalty = -139.97
- LanguageModel = +327.98
- LanguageModel_OOV = -234.7
- PhraseModel_0 = -144.49
- PhraseModel_1 = -263.88
- PhraseModel_2 = -149.25
- PhraseModel_3 = -38.805
- PhraseModel_4 = +50.575
- PhraseModel_5 = -52.4
- PhraseModel_6 = +41.6
- PassThrough = -230.2
+ Glue = -44.3
+ WordPenalty = -131.85
+ LanguageModel = +230.91
+ LanguageModel_OOV = -285.4
+ PhraseModel_0 = -194.27
+ PhraseModel_1 = -294.83
+ PhraseModel_2 = -92.043
+ PhraseModel_3 = -140.24
+ PhraseModel_4 = +85.613
+ PhraseModel_5 = +238.1
+ PhraseModel_6 = +158.7
+ PassThrough = -359.6
---
- 1best avg score: 0.36222 (+0.094717)
- 1best avg model score: -17416 (-7355.5)
- avg # pairs: 661.2
+ 1best avg score: 0.37375 (+0.078067)
+ 1best avg model score: -14519 (+1359.7)
+ avg # pairs: 545.4
avg # rank err: 0 (meaningless)
avg # margin viol: 0
k-best loss imp: 100%
- non0 feature count: 2163
+ non0 feature count: 2218
avg list sz: 91.3
- avg f count: 132.53
-(time 0.33 min, 2 s/S)
+ avg f count: 137.77
+(time 0.35 min, 2.1 s/S)
Writing weights file to '-' ...
done
---
-Best iteration: 3 [SCORE 'fixed_stupid_bleu'=0.36222].
-This took 1.0833 min.
+Best iteration: 3 [SCORE 'fixed_stupid_bleu'=0.37375].
+This took 1.05 min.
diff --git a/training/dtrain/examples/toy/cdec.ini b/training/dtrain/examples/toy/cdec.ini
index b14f4819..e6c19abe 100644
--- a/training/dtrain/examples/toy/cdec.ini
+++ b/training/dtrain/examples/toy/cdec.ini
@@ -1,3 +1,4 @@
formalism=scfg
add_pass_through_rules=true
grammar=grammar.gz
+#add_extra_pass_through_features=6
diff --git a/training/dtrain/examples/toy/dtrain.ini b/training/dtrain/examples/toy/dtrain.ini
index cd715f26..ef956df7 100644
--- a/training/dtrain/examples/toy/dtrain.ini
+++ b/training/dtrain/examples/toy/dtrain.ini
@@ -2,7 +2,7 @@ decoder_config=cdec.ini
input=src
refs=tgt
output=-
-print_weights=logp shell_rule house_rule small_rule little_rule PassThrough
+print_weights=logp shell_rule house_rule small_rule little_rule PassThrough PassThrough_1 PassThrough_2 PassThrough_3 PassThrough_4 PassThrough_5 PassThrough_6
k=4
N=4
epochs=2
diff --git a/training/dtrain/parallelize.rb b/training/dtrain/parallelize.rb
index 60ca9422..5fc8b04e 100755
--- a/training/dtrain/parallelize.rb
+++ b/training/dtrain/parallelize.rb
@@ -4,7 +4,7 @@ require 'trollop'
def usage
STDERR.write "Usage: "
- STDERR.write "ruby parallelize.rb -c <dtrain.ini> [-e <epochs=10>] [--randomize/-z] [--reshard/-y] -s <#shards|0> [-p <at once=9999>] -i <input> -r <refs> [--qsub/-q] [--dtrain_binary <path to dtrain binary>] [-l \"l2 select_k 100000\"] [--extra_qsub \"-l virtual_free=24G\"]\n"
+ STDERR.write "ruby parallelize.rb -c <dtrain.ini> [-e <epochs=10>] [--randomize/-z] [--reshard/-y] -s <#shards|0> [-p <at once=9999>] -i <input> -r <refs> [--qsub/-q] [--dtrain_binary <path to dtrain binary>] [-l \"l2 select_k 100000\"] [--extra_qsub \"-l mem_free=24G\"]\n"
exit 1
end
@@ -26,7 +26,6 @@ opts = Trollop::options do
end
usage if not opts[:config]&&opts[:shards]&&opts[:input]&&opts[:references]
-
dtrain_dir = File.expand_path File.dirname(__FILE__)
if not opts[:dtrain_binary]
dtrain_bin = "#{dtrain_dir}/dtrain"
@@ -56,6 +55,7 @@ refs = opts[:references]
use_qsub = opts[:qsub]
shards_at_once = opts[:processes_at_once]
first_input_weights = opts[:first_input_weights]
+opts[:extra_qsub] = "-l #{opts[:extra_qsub]}" if opts[:extra_qsub]!=""
`mkdir work`
@@ -64,8 +64,9 @@ def make_shards(input, refs, num_shards, epoch, rand)
index = (0..lc-1).to_a
index.reverse!
index.shuffle! if rand
- shard_sz = lc / num_shards
- leftover = lc % num_shards
+ shard_sz = (lc / num_shards.to_f).round 0
+ leftover = lc - (num_shards*shard_sz)
+ leftover = 0 if leftover < 0
in_f = File.new input, 'r'
in_lines = in_f.readlines
refs_f = File.new refs, 'r'
@@ -74,7 +75,10 @@ def make_shards(input, refs, num_shards, epoch, rand)
shard_refs_files = []
in_fns = []
refs_fns = []
+ new_num_shards = 0
0.upto(num_shards-1) { |shard|
+ break if index.size==0
+ new_num_shards += 1
in_fn = "work/shard.#{shard}.#{epoch}.in"
shard_in = File.new in_fn, 'w+'
in_fns << in_fn
@@ -83,6 +87,7 @@ def make_shards(input, refs, num_shards, epoch, rand)
refs_fns << refs_fn
0.upto(shard_sz-1) { |i|
j = index.pop
+ break if !j
shard_in.write in_lines[j]
shard_refs.write refs_lines[j]
}
@@ -98,7 +103,7 @@ def make_shards(input, refs, num_shards, epoch, rand)
(shard_in_files + shard_refs_files).each do |f| f.close end
in_f.close
refs_f.close
- return [in_fns, refs_fns]
+ return in_fns, refs_fns, new_num_shards
end
input_files = []
@@ -111,7 +116,7 @@ if predefined_shards
end
num_shards = input_files.size
else
- input_files, refs_files = make_shards input, refs, num_shards, 0, rand
+ input_files, refs_files, num_shards = make_shards input, refs, num_shards, 0, rand
end
0.upto(epochs-1) { |epoch|
@@ -158,7 +163,7 @@ end
`#{cat} work/weights.*.#{epoch} > work/weights_cat`
`#{ruby} #{lplp_rb} #{lplp_args} #{num_shards} < work/weights_cat > work/weights.#{epoch}`
if rand and reshard and epoch+1!=epochs
- input_files, refs_files = make_shards input, refs, num_shards, epoch+1, rand
+ input_files, refs_files, num_shards = make_shards input, refs, num_shards, epoch+1, rand
end
}
diff --git a/training/mira/kbest_cut_mira.cc b/training/mira/kbest_cut_mira.cc
index 56206593..724b1853 100644
--- a/training/mira/kbest_cut_mira.cc
+++ b/training/mira/kbest_cut_mira.cc
@@ -95,7 +95,8 @@ bool InitCommandLine(int argc, char** argv, po::variables_map* conf) {
("stream,t", "Stream mode (used for realtime)")
("weights_output,O",po::value<string>(),"Directory to write weights to")
("output_dir,D",po::value<string>(),"Directory to place output in")
- ("decoder_config,c",po::value<string>(),"Decoder configuration file");
+ ("decoder_config,c",po::value<string>(),"Decoder configuration file")
+ ("verbose,v",po::value<bool>()->zero_tokens(),"verbose stderr output");
po::options_description clo("Command line options");
clo.add_options()
("config", po::value<string>(), "Configuration file")
@@ -621,6 +622,7 @@ int main(int argc, char** argv) {
vector<string> corpus;
+ const bool VERBOSE = conf.count("verbose");
const string metric_name = conf["mt_metric"].as<string>();
optimizer = conf["optimizer"].as<int>();
fear_select = conf["fear"].as<int>();
@@ -783,7 +785,8 @@ int main(int argc, char** argv) {
double margin = cur_bad.features.dot(dense_weights) - cur_good.features.dot(dense_weights);
double mt_loss = (cur_good.mt_metric - cur_bad.mt_metric);
const double loss = margin + mt_loss;
- cerr << "LOSS: " << loss << " Margin:" << margin << " BLEUL:" << mt_loss << " " << cur_bad.features.dot(dense_weights) << " " << cur_good.features.dot(dense_weights) <<endl;
+ cerr << "LOSS: " << loss << " Margin:" << margin << " BLEUL:" << mt_loss << endl;
+ if (VERBOSE) cerr << cur_bad.features.dot(dense_weights) << " " << cur_good.features.dot(dense_weights) << endl;
if (loss > 0.0 || !checkloss) {
SparseVector<double> diff = cur_good.features;
diff -= cur_bad.features;
@@ -920,6 +923,7 @@ int main(int argc, char** argv) {
lambdas += (cur_pair[1]->features) * step_size;
lambdas -= (cur_pair[0]->features) * step_size;
+ if (VERBOSE) cerr << " Lambdas " << lambdas << endl;
//reload weights based on update
dense_weights.clear();
diff --git a/training/mira/mira.py b/training/mira/mira.py
index 3e6aa2db..ec9c2d64 100755
--- a/training/mira/mira.py
+++ b/training/mira/mira.py
@@ -143,6 +143,12 @@ def main():
parser.add_argument('--pass-suffix',
help='multipass decoding iteration. see documentation '
'at www.cdec-decoder.org for more information')
+ parser.add_argument('--qsub',
+ help='use qsub', action='store_true')
+ parser.add_argument('--pmem',
+ help='memory for qsub', type=str, default='5G')
+ parser.add_argument('-v', '--verbose',
+ help='more verbose mira optimizers')
args = parser.parse_args()
args.metric = args.metric.upper()
@@ -315,6 +321,8 @@ def split_devset(dev, outdir):
def optimize(args, script_dir, dev_size):
parallelize = script_dir+'/../utils/parallelize.pl'
+ if args.qsub:
+ parallelize += " -p %s"%args.pmem
decoder = script_dir+'/kbest_cut_mira'
(source, refs) = split_devset(args.devset, args.output_dir)
port = random.randint(15000,50000)
@@ -353,10 +361,15 @@ def optimize(args, script_dir, dev_size):
decoder_cmd += ' -a'
if not args.no_pseudo:
decoder_cmd += ' -e'
+ if args.verbose:
+ decoder_cmd += ' -v'
- #always use fork
- parallel_cmd = '{0} --use-fork -e {1} -j {2} --'.format(
- parallelize, logdir, args.jobs)
+ if args.qsub:
+ parallel_cmd = '{0} -e {1} -j {2} --'.format(
+ parallelize, logdir, args.jobs)
+ else:
+ parallel_cmd = '{0} --use-fork -e {1} -j {2} --'.format(
+ parallelize, logdir, args.jobs)
cmd = parallel_cmd + ' ' + decoder_cmd
logging.info('OPTIMIZATION COMMAND: {}'.format(cmd))
diff --git a/training/pro/pro.pl b/training/pro/pro.pl
index 3b30c379..8ebb5864 100755
--- a/training/pro/pro.pl
+++ b/training/pro/pro.pl
@@ -69,17 +69,19 @@ my $reg_previous = 5000;
# Process command-line options
if (GetOptions(
- "config=s" => \$iniFile,
- "weights=s" => \$initial_weights,
- "devset=s" => \$devset,
- "jobs=i" => \$jobs,
- "metric=s" => \$metric,
- "pass-suffix=s" => \$pass_suffix,
- "qsub" => \$useqsub,
- "help" => \$help,
- "reg=f" => \$reg,
- "reg-previous=f" => \$reg_previous,
- "output-dir=s" => \$dir,
+ "config=s" => \$iniFile,
+ "weights=s" => \$initial_weights,
+ "devset=s" => \$devset,
+ "jobs=i" => \$jobs,
+ "max-iterations=i" => \$max_iterations,
+ "metric=s" => \$metric,
+ "pass-suffix=s" => \$pass_suffix,
+ "qsub" => \$useqsub,
+ "help" => \$help,
+ "reg=f" => \$reg,
+ "reg-previous=f" => \$reg_previous,
+ "pmem=s" => \$pmem,
+ "output-dir=s" => \$dir,
) == 0 || @ARGV!=0 || $help) {
print_help();
exit;
generated by cgit v1.2.3 (git 2.39.1) at 2024-12-30 19:59:33 +0000