Added NDEBUG, functions for merging models.

Author: tmbdev

SConstruct

@@ -77,7 +77,7 @@ elif option("debug", 0)>0:
     env.Append(CCFLAGS="-g".split())
     env.Append(LINKFLAGS="-g".split())
 else:
-    env.Append(CXXFLAGS="-g -O3 -finline".split())
+    env.Append(CXXFLAGS="-g -O3 -DNDEBUG -finline".split())
     env.Append(CCFLAGS="-g".split())
 
 # Extra layers (old layers or testing)

clstm.cc

@@ -540,12 +540,25 @@ static void get_allparams(vector<vector<Params*>> &allparams, vector<Network> &n
   }
 }
 
-void share_deltas(vector<Network> networks) {
+void distribute_weights(vector<Network> &networks, int from) {
   vector<vector<Params*>> allparams;
   get_allparams(allparams, networks);
   int n = allparams.size();
   int m = allparams[0].size();
-  for(int i=1; n; i++) {
+  for(int i=0; i<n; i++) {
+    if (i==from) continue;
+    for(int j=0; j<m; j++) {
+      allparams[i][j]->V() = allparams[from][j]->V();
+    }
+  }
+}
+
+void share_deltas(vector<Network> &networks) {
+  vector<vector<Params*>> allparams;
+  get_allparams(allparams, networks);
+  int n = allparams.size();
+  int m = allparams[0].size();
+  for(int i=1; i<n; i++) {
     for(int j=0; j<m; j++) {
       allparams[0][j]->D() += allparams[i][j]->D();
     }
@@ -555,7 +568,7 @@ void share_deltas(vector<Network> networks) {
   }
 }
 
-void average_weights(vector<Network> networks) {
+void average_weights(vector<Network> &networks) {
   vector<vector<Params*>> allparams;
   get_allparams(allparams, networks);
   int n = allparams.size();
@@ -564,13 +577,11 @@ void average_weights(vector<Network> networks) {
     for(int j=0; j<m; j++) {
       allparams[0][j]->V() += allparams[i][j]->V();
     }
-    for(int j=0; j<m; j++) {
-      allparams[0][j]->V() = allparams[0][j]->V() * Float(1.0/n);
-    }
-    for(int j=0; j<m; j++) {
-      allparams[i][j]->V() = allparams[0][j]->V();
-    }
   }
+  for(int j=0; j<m; j++) {
+    allparams[0][j]->V() = allparams[0][j]->V() * Float(1.0/n);
+  }
+  distribute_weights(networks);
 }
 
 }  // namespace ocropus

clstm.h

@@ -205,8 +205,9 @@ void trivial_decode(Classes &cs, Sequence &outputs, int batch,
 // single sequence training functions
 void mktargets(Tensor<float,2> &seq, Tensor<int,1> &targets, int ndim);
 
-void share_deltas(vector<Network> networks);
-void average_weights(vector<Network> networks);
+void share_deltas(vector<Network> &networks);
+void average_weights(vector<Network> &networks);
+void distribute_weights(vector<Network> &networks, int from=0);
 }
 
 namespace {

clstmhl.h

@@ -108,7 +108,7 @@ struct CLSTMText {
 };
 
 struct CLSTMOCR {
-  unique_ptr<INormalizer> normalizer;
+  shared_ptr<INormalizer> normalizer;
   Network net;
   int target_height = 48;
   int nclasses = -1;
@@ -132,7 +132,7 @@ struct CLSTMOCR {
     normalizer.reset(make_CenterNormalizer());
     normalizer->target_height = target_height;
   }
-  std::wstring train(Tensor<float,2> &raw, const std::wstring &target) {
+  std::wstring fwdbwd(Tensor<float,2> &raw, const std::wstring &target) {
     normalizer->measure(raw);
     normalizer->normalize(image, raw);
     set_inputs(net, image);
@@ -144,11 +144,18 @@ struct CLSTMOCR {
     for (int t = 0; t < aligned.size(); t++)
       net->outputs[t].D() = aligned[t].V() - net->outputs[t].V();
     net->backward();
-    sgd_update(net);
     Classes outputs;
     trivial_decode(outputs, net->outputs);
     return net->codec.decode(outputs);
   }
+  void update() {
+    sgd_update(net);
+  }
+  std::wstring train(Tensor<float,2> &raw, const std::wstring &target) {
+    std::wstring result = fwdbwd(raw, target);
+    update();
+    return result;
+  }
   std::string aligned_utf8() {
     Classes outputs;
     trivial_decode(outputs, aligned);

run-tests

@@ -4,7 +4,7 @@ export PS4='
 >>>>>>> '
 trap "echo TEST FAILED" EXIT
 set -x
-export seed=0.1423
+export seed=0.7733
 scons -s -c; rm -f *.o *.a
 scons -j 4 clstmocrtrain clstmfiltertrain clstmfilter clstmocr test-lstm
 ./test-lstm