initiall cluster parallel CG code

1 files changed, 15 insertions, 12 deletions

diff --git a/cg.cc b/cg.cc
index 0ca2bceb..b7c7bf33 100644
--- a/cg.cc
+++ b/cg.cc
@@ -1,10 +1,10 @@
 /*
-Copyright (c) 2009 Yahoo! Inc.  All rights reserved.  The copyrights
+Copyright (c) 2009-2011 Yahoo! Inc.  All rights reserved.  The copyrights
 embodied in the content of this file are licensed under the BSD
 (revised) open source license
 
-The algorithm here is generally based on Jonathan Shewchuck's tutorial.
-
+The algorithm here is generally based on Jonathan Shewchuck's
+tutorial.  
  */
 #include <fstream>
 #include <float.h>
@@ -342,7 +342,10 @@ void setup_cg(gd_thread_params t)
 	      if (current_pass == 0)
 		{
 		  if(global.master_location != "")
-		    accumulate(socks, reg, 3); //Accumulate preconditioner
+		    {
+		      accumulate(socks, reg, 3); //Accumulate preconditioner
+		      importance_weight_sum = accumulate_scalar(socks, importance_weight_sum);
+		    }
 		  finalize_preconditioner(reg,global.regularization);
 		}
 	      if (gradient_pass) // We just finished computing all gradients
@@ -354,7 +357,7 @@ void setup_cg(gd_thread_params t)
 		  if (global.regularization > 0.)
 		    loss_sum += add_regularization(reg,global.regularization);
 		  if (!global.quiet)
-		    fprintf(stderr, "%-f\t", loss_sum );
+		    fprintf(stderr, "%-f\t", loss_sum / importance_weight_sum);
 		  
 		  if (current_pass > 0 && loss_sum > previous_loss_sum)
 		    {// we stepped to far last time, step back
@@ -380,7 +383,7 @@ void setup_cg(gd_thread_params t)
 		      float new_d_mag = derivative_magnitude(reg, old_first_derivative);
 		      previous_d_mag = new_d_mag;
 		      if (!global.quiet)
-			fprintf(stderr, "%f\t%f\t", mix_frac, new_d_mag);
+			fprintf(stderr, "%f\t%f\t", mix_frac, new_d_mag / importance_weight_sum);
 		      
 		      update_direction(reg, mix_frac, old_first_derivative);
 		      gradient_pass = false;//now start computing curvature
@@ -402,7 +405,7 @@ void setup_cg(gd_thread_params t)
 		    }
 		  step_size = - dd/curvature;
 		  if (!global.quiet) {
-		    fprintf(stderr, "%-e\t%-e\t%-e\t%-f\n", curvature, d_mag, step_size, 0.5*step_size*step_size*curvature);
+		    fprintf(stderr, "%-e\t%-e\t%-e\t%-f\n", curvature/importance_weight_sum, d_mag/importance_weight_sum, step_size, 0.5*step_size*step_size*curvature /importance_weight_sum);
 		    //fprintf(stdout, "Net comm. time is %f\n",net_comm_time - prev_comm_time);
 		  }
 		  prev_comm_time = net_comm_time;
@@ -480,8 +483,11 @@ void setup_cg(gd_thread_params t)
 	    }
 	  ftime(&t_end_global);
 	  net_time += (int) (1000.0 * (t_end_global.time - t_start_global.time) + (t_end_global.millitm - t_start_global.millitm)); 
-	  cerr<<"Net time spent in communication = "<<(float)net_comm_time/(float)1000<<"seconds\n";
-	  cerr<<"Net time spent = "<<(float)net_time/(float)1000<<"seconds\n";
+	  if (!global.quiet)
+	    {
+	      cerr<<"Net time spent in communication = "<<(float)net_comm_time/(float)1000<<" seconds\n";
+	      cerr<<"Net time spent = "<<(float)net_time/(float)1000<<" seconds\n";
+	    }
 	  if (global.local_prediction > 0)
 	    shutdown(global.local_prediction, SHUT_WR);
 	  if(global.master_location != "")
@@ -495,13 +501,10 @@ void setup_cg(gd_thread_params t)
 	;//busywait when we have predicted on all examples but not yet trained on all.
     }
   
-  cerr<<"Done CG\n";
-  fflush(stderr);
   if(global.master_location != "")
     all_reduce_close(socks);
   free(predictions.begin);
   free(old_first_derivative);
-  cerr<<"Really Done CG\n";
   ftime(&t_end_global);
   net_time += (int) (1000.0 * (t_end_global.time - t_start_global.time) + (t_end_global.millitm - t_start_global.millitm)); 
   cerr<<"Net time spent in communication = "<<(float)net_comm_time/(float)1000<<"seconds\n";