Added new param lda_epsilon to replace the hard-coded convergence threshold used in the lda_loop function.

3 files changed, 4 insertions, 1 deletions

diff --git a/vowpalwabbit/global_data.cc b/vowpalwabbit/global_data.cc
index 844e6db3..a0d602ee 100644
--- a/vowpalwabbit/global_data.cc
+++ b/vowpalwabbit/global_data.cc
@@ -238,6 +238,7 @@ vw::vw()
   lda_alpha = 0.1f;
   lda_rho = 0.1f;
   lda_D = 10000.;
+  lda_epsilon = 0.001;
   minibatch = 1;
   span_server = "";
   m = 15;
diff --git a/vowpalwabbit/global_data.h b/vowpalwabbit/global_data.h
index 0b78ef42..40e12718 100644
--- a/vowpalwabbit/global_data.h
+++ b/vowpalwabbit/global_data.h
@@ -220,6 +220,7 @@ struct vw {
   float lda_alpha;
   float lda_rho;
   float lda_D;
+  float lda_epsilon;
 
   std::string text_regressor_name;
   std::string inv_hash_regressor_name;
diff --git a/vowpalwabbit/lda_core.cc b/vowpalwabbit/lda_core.cc
index c7be0d9f..db2281a2 100644
--- a/vowpalwabbit/lda_core.cc
+++ b/vowpalwabbit/lda_core.cc
@@ -495,7 +495,7 @@ v_array<float> old_gamma;
       for (size_t k =0; k<all.lda; k++)

 	new_gamma[k] = new_gamma[k]*v[k]+all.lda_alpha;

     }

-  while (average_diff(all, old_gamma.begin, new_gamma.begin) > 0.001);

+  while (average_diff(all, old_gamma.begin, new_gamma.begin) > all.lda_epsilon);

 

   ec->topic_predictions.erase();

   ec->topic_predictions.resize(all.lda);

@@ -748,6 +748,7 @@ learner* setup(vw&all, std::vector<std::string>&opts, po::variables_map& vm)
     ("lda_alpha", po::value<float>(&all.lda_alpha), "Prior on sparsity of per-document topic weights")

     ("lda_rho", po::value<float>(&all.lda_rho), "Prior on sparsity of topic distributions")

     ("lda_D", po::value<float>(&all.lda_D), "Number of documents")

+    ("lda_epsilon", po::value<float>(&all.lda_epsilon), "Loop convergence threshold")

     ("minibatch", po::value<size_t>(&all.minibatch), "Minibatch size, for LDA");

 

   po::parsed_options parsed = po::command_line_parser(opts).