From 2b9bc7c2f2a4a3993196a0a3f89f4007143ad351 Mon Sep 17 00:00:00 2001
From: redpony <redpony@ec762483-ff6d-05da-a07a-a48fb63a330f>
Date: Fri, 15 Oct 2010 20:30:49 +0000
Subject: few fixes

git-svn-id: https://ws10smt.googlecode.com/svn/trunk@676 ec762483-ff6d-05da-a07a-a48fb63a330f
---
 training/mpi_online_optimize.cc | 13 ++++++++-----
 1 file changed, 8 insertions(+), 5 deletions(-)

(limited to 'training')

diff --git a/training/mpi_online_optimize.cc b/training/mpi_online_optimize.cc
index 4c08b181..96f3bcfb 100644
--- a/training/mpi_online_optimize.cc
+++ b/training/mpi_online_optimize.cc
@@ -220,7 +220,6 @@ bool LoadAgenda(const string& file, vector<pair<string, int> >* a) {
     x.first = line.substr(0,d);
     x.second = atoi(line.substr(d+1).c_str());
     a->push_back(x);
-    cerr << "X: " << x.second << " - " << x.first << "'\n";
     if (!FileExists(x.first)) {
       cerr << "Can't find file " << x.first << endl;
       return false;
@@ -261,7 +260,11 @@ int main(int argc, char** argv) {
     return 1;
   }
 
+  size_t total_corpus_size = 0;
+  reduce(world, corpus.size(), total_corpus_size, std::plus<size_t>(), 0);
+
   if (rank == 0) {
+    cerr << "Total corpus size: " << total_corpus_size << endl;
     const unsigned batch_size = size_per_proc * size;
     // TODO config
     lr.reset(new ExponentialDecayLearningRate(batch_size, conf["eta_0"].as<double>()));
@@ -269,7 +272,7 @@ int main(int argc, char** argv) {
     const string omethod = conf["optimization_method"].as<string>();
     if (omethod == "sgd") {
       const double C = conf["regularization_strength"].as<double>();
-      o.reset(new CumulativeL1OnlineOptimizer(lr, corpus.size(), C));
+      o.reset(new CumulativeL1OnlineOptimizer(lr, total_corpus_size, C));
     } else {
       assert(!"fail");
     }
@@ -302,7 +305,8 @@ int main(int argc, char** argv) {
     ReadFile ini_rf(cur_config);
     Decoder decoder(ini_rf.stream());
 
-    o->ResetEpoch(); // resets the learning rate-- TODO is this good?
+    if (rank == 0)
+      o->ResetEpoch(); // resets the learning rate-- TODO is this good?
 
     int iter = -1;
     bool converged = false;
@@ -324,7 +328,7 @@ int main(int argc, char** argv) {
         }
         if (converged && ((ai+1)==agenda.size())) { fname = "weights.final.gz"; }
         ostringstream vv;
-        vv << "total iter=" << titer << " (of current config iter=" << iter << ")  minibatch=" << size_per_proc << " sentences/proc x " << size << " procs.   num_feats=" << x.size() << '/' << FD::NumFeats() << "   passes_thru_data=" << (titer * size * size_per_proc / static_cast<double>(corpus.size())) << "   eta=" << lr->eta(titer);
+        vv << "total iter=" << titer << " (of current config iter=" << iter << ")  minibatch=" << size_per_proc << " sentences/proc x " << size << " procs.   num_feats=" << x.size() << '/' << FD::NumFeats() << "   passes_thru_data=" << (titer * size_per_proc / static_cast<double>(corpus.size())) << "   eta=" << lr->eta(titer);
         const string svv = vv.str();
         cerr << svv << endl;
         weights.WriteToFile(fname, true, &svv);
@@ -343,7 +347,6 @@ int main(int argc, char** argv) {
       if (rank == 0) {
         g /= (size_per_proc * size);
         o->UpdateWeights(g, FD::NumFeats(), &x);
-        cerr << "XX: " << x << endl;
       }
       broadcast(world, x, 0);
       broadcast(world, converged, 0);
-- 
cgit v1.2.3