add label smoothing

source/sample/transformer/T2TTrainer.cpp

@@ -108,6 +108,7 @@ void T2TTrainer::Init(int argc, char ** argv)
     LoadParamFloat(argc, argv, "adambeta2", &adamBeta2, 0.999F);
     LoadParamFloat(argc, argv, "adamdelta", &adamDelta, 1e-8F);
     LoadParamBool(argc, argv, "shuffled", &isShuffled, false);
+    LoadParamFloat(argc, argv, "labelsmoothing", &labelSmoothingP, 0);
     LoadParamInt(argc, argv, "nstepcheckpoint", &nStepCheckpoint, -1);
     LoadParamBool(argc, argv, "epochcheckpoint", &useEpochCheckpoint, false);
 
@@ -180,6 +181,9 @@ void T2TTrainer::Train(const char * fn, const char * validFN, const char * model
         /* gold standard */
         XTensor gold;
         
+        /* label smoothed gold standard (if needed) */
+        XTensor goldSmoothed;
+        
         while(LoadBatch(file, true, &batch, &padding, &gold, NULL, 1, vSize, sBatchSize, wBatchSize, isLenSorted, wc, devID, mem)){
 
             CheckNTErrors(batch.order == 3, "wrong tensor order of the sequence batch");
@@ -190,12 +194,17 @@ void T2TTrainer::Train(const char * fn, const char * validFN, const char * model
             /* make the network */
             model->Make(batch, output, padding, true);
 
+            /* back-propagation for obtaining gradients */
+            if(labelSmoothingP > 0)
+                LabelSmooth(&gold, &goldSmoothed, labelSmoothingP);
+            
             /* make paddings for the output */
             if(output.GetDim(0) > 1)
-                PadOutput(&output, &padding);
+                PadOutput(&output, &gold, &padding);
             
-            /* back-propagation for obtaining gradients */
-            net.Backward(output, gold, CROSSENTROPY);
+            XTensor &g = labelSmoothingP > 0 ? goldSmoothed : gold;
+            
+            net.Backward(output, g, CROSSENTROPY);
             
             /* learning rate */
             lr = lrate * (1.0F / (float)sqrt((float)d)) * (float)MIN(pow((float)step + 1, -0.5F - lrbias), ((float)step + 1) * pow((float)nwarmup, -1.5F - lrbias));
@@ -789,9 +798,11 @@ void T2TTrainer::PrepareModel(T2TModel * model)
 /* 
 do padding on the output 
 >> output - output tensor of the network
+>> gold - gold standard
 >> padding - padding of a batch of sentences
+>> lsP - smoothing factor
 */
-void T2TTrainer::PadOutput(XTensor * output, XTensor * padding)
+void T2TTrainer::PadOutput(XTensor * output, XTensor * gold, XTensor * padding)
 {
     if(output == NULL || padding == NULL)
         return;
@@ -807,13 +818,45 @@ void T2TTrainer::PadOutput(XTensor * output, XTensor * padding)
 
     _CopyValues(padding, padding2);
     _ScaleAndShiftMe(padding2, 1e9F, -1e9F);
-
     _SumDim(output, padding2, output, 0);
     
     output->Reshape(on, dimso);
     
+    if(gold != NULL){
+        gold->Reshape(gold->unitNum/dimso[output->order - 1], dimso[output->order - 1]);
+        _CopyValues(padding, padding2);
+        _MultiplyDim(gold, padding2, gold, 0);
+        gold->Reshape(on, dimso);
+    }
+
     delete[] dimso;
     DelTensorBuf(padding2);
 }
     
+/*
+perform label smoothing
+>> gold - gold standard
+>> smoothed - result of label smoothing
+>> lsP - smoothing factor
+*/
+void T2TTrainer::LabelSmooth(XTensor * gold, XTensor * smoothed, DTYPE lsP)
+{
+    DTYPE p = lsP;
+    CheckNTErrors(p >= 0 && p <= 1.0F, "Smoothing factor must be in range [0,1]");
+    
+    int n = gold->GetDim(-1);
+    DTYPE q = 1.0F - p;
+    DTYPE gift = p / (n - 1);
+    
+    InitTensor(smoothed, gold);
+    _CopyValues(gold, smoothed);
+    
+    if(p == 0)
+        return;
+    
+    _ScaleAndShiftMe(smoothed, gift/q, -gift/q);
+    _Sum(smoothed, gold, smoothed);
+    _ScaleAndShiftMe(smoothed, q);
+}
+
 }

source/sample/transformer/T2TTrainer.h

@@ -116,6 +116,9 @@ public:
     /* indicates whether the data file is shuffled for training */
     bool isShuffled;
     
+    /* the factor of label smoothing */
+    DTYPE labelSmoothingP;
+
     /* number of steps after which we make a checkpoint */
     int nStepCheckpoint;
 
@@ -168,7 +171,10 @@ public:
     void PrepareModel(T2TModel * model);
 
     /* do padding on the output */
-    void PadOutput(XTensor * output, XTensor * padding);
+    void PadOutput(XTensor * output, XTensor * gold, XTensor * padding);
+    
+    /* perform label smoothing */
+    void LabelSmooth(XTensor * gold, XTensor * smoothed, DTYPE lsP);
 };
 
 

source/sample/transformer/Transformer.cpp

@@ -19,6 +19,7 @@
  * $Created by: XIAO Tong (xiaotong@mail.neu.edu.cn) 2018-07-31
  */
 
+#include <math.h>
 #include "Transformer.h"
 #include "T2TModel.h"
 #include "T2TUtility.h"
@@ -33,6 +34,8 @@ int TransformerMain(int argc, const char ** argv)
     if(argc == 0)
         return 1;
     
+    fprintf(stderr, "%e\n", log(1e-45F));
+
     char ** args = new char*[argc];
     for(int i = 0; i < argc; i++){
         args[i] = new char[strlen(argv[i]) + 1];