ealdent-lda-ruby 0.1.6 → 0.1.7

data/README

@@ -0,0 +1,22 @@
+Latent Dirichlet Allocation – Ruby Wrapper
+
+This wrapper is based on C-code by David M. Blei. In a nutshell, it can be used to automatic cluster documents into topics. The number of topics are chosen beforehand and the topics found are usually fairly intuitive. Details of the implementation can be found in the paper by Blei, Ng, and Jordan.
+
+The original C code relied on files for the input and output. We felt it was necessary to depart from that model and use Ruby objects for these steps instead. The only file necessary will be the data file (in a format similar to that used by SVMlight). Optionally you may need a vocabulary file to be able to extract the words belonging to topics.
+
+Example usage:
+
+require 'lda'
+lda = Lda::Lda.new      # create an Lda object for training
+corpus = Lda::Corpus.new("data/data_file.dat")
+lda.corpus = corpus
+lda.em("random")        # run EM algorithm using random starting points
+lda.load_vocabulary("data/vocab.txt")
+lda.print_topics(20)    # print the topic 20 words per topic
+
+See the rdocs for further information. You can also check out the mailing list for this project if you have any questions or mail lda-ruby@groups.google.com [email link]. If you have general questions about Latent Dirichlet Allocation, I urge you to use the topic models mailing list, since the people who monitor that are very knowledgeable.
+
+
+References
+
+Blei, David M., Ng, Andrew Y., and Jordan, Michael I. 2003. Latent dirichlet allocation. Journal of Machine Learning Research. 3 (Mar. 2003), 993-1022. 

data/lib/lda-inference.c

@@ -837,27 +837,44 @@ static VALUE wrap_get_gamma(VALUE self) {
 	return arr;
 }
 
+
+/*
+ * Compute the phi values by running inference after the initial EM run has been completed.
+ *
+ * Returns a 3D matrix:  <tt>num_docs x length x num_topics</tt>.
+ */
 static VALUE wrap_get_phi(VALUE self) {
     if (!model_loaded)
         return Qnil;
     
-    VALUE arr;
-    int i = 0, j = 0;
-    int max_length = max_corpus_length(last_corpus);
+    VALUE arr = rb_ary_new2(last_corpus->num_docs);
+    int i = 0, j = 0, k = 0;
     
+    int max_length = max_corpus_length(last_corpus);
     
-    arr = rb_ary_new2(max_length);
-    for (i = 0; i < max_length; i++) {
-        VALUE arr2 = rb_ary_new2(last_model->num_topics);
-        for (j = 0; j < last_model->num_topics; j++) {
-            rb_ary_store(arr2, j, rb_float_new(last_phi[i][j]));
+    for (i = 0; i < last_corpus->num_docs; i++) {
+        VALUE arr1 = rb_ary_new2(last_corpus->docs[i].length);
+        
+        lda_inference(&(last_corpus->docs[i]), last_model, last_gamma[i], last_phi);
+        
+        for (j = 0; j < last_corpus->docs[i].length; j++) {
+            VALUE arr2 = rb_ary_new2(last_model->num_topics);
+            
+            for (k = 0; k < last_model->num_topics; k++) {
+                rb_ary_store(arr2, k, rb_float_new(last_phi[j][k]));
+            }
+            
+            rb_ary_store(arr1, j, arr2);
         }
-        rb_ary_store(arr, i, arr2);
+        
+        rb_ary_store(arr, i, arr1);
     }
     
     return arr;
 }
 
+
+
 /*
  * Get the beta matrix after the model has been run.
  */
@@ -963,7 +980,7 @@ void Init_lda_ext() {
 	// retrieve model and gamma
 	rb_define_method(rb_cLda, "beta", wrap_get_model_beta, 0);
 	rb_define_method(rb_cLda, "gamma", wrap_get_gamma, 0);
-    rb_define_method(rb_cLda, "phi", wrap_get_phi, 0);
+    rb_define_method(rb_cLda, "compute_phi", wrap_get_phi, 0);
 	rb_define_method(rb_cLda, "model", wrap_get_model_settings, 0);
 }
 

data/lib/lda.rb

@@ -114,6 +114,7 @@ module Lda
       self.load_default_settings
       @corpus = nil
       @vocab = nil
+      @phi = nil
     end
     
     # 
@@ -248,6 +249,49 @@ module Lda
       topics
     end
     
+    
+    #
+    # Get the phi matrix which can be used to assign probabilities to words
+    # belonging to a specific topic in each document.  The return value is a
+    # 3D matrix:  num_docs x doc_length x num_topics.  The value is cached
+    # after the first call, so if it needs to be recomputed, set the +recompute+
+    # value to true.
+    #
+    def phi(recompute=false)
+      if not @phi or recompute
+        # either the phi variable has not been instantiated or the recompute flag has been set
+        @phi = self.compute_phi
+      end
+      @phi
+    end
+    
+    
+    
+    #
+    # Compute the average log probability for each topic for each document in the corpus.
+    # This method returns a matrix:  num_docs x num_topics with the average log probability
+    # for the topic in the document.
+    #
+    def compute_topic_document_probability
+      outp = Array.new
+      @corpus.documents.each_with_index do |doc, idx|
+        tops = [0.0] * self.num_topics
+        ttl  = doc.counts.inject(0.0) {|sum, i| sum + i}
+        self.phi[idx].each_with_index do |word_dist, word_idx|
+          word_dist.each_with_index do |top_prob, top_idx|
+            tops[top_idx] += Math.log(top_prob) * doc.counts[word_idx]
+          end
+        end
+        tops = tops.map {|i| i / ttl}
+        outp << tops
+      end
+      
+      outp
+    end
+    
+    
+    
+    
     #
     # String representation displaying current settings.
     #

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification 
 name: ealdent-lda-ruby
 version: !ruby/object:Gem::Version 
-  version: 0.1.6
+  version: 0.1.7
 platform: ruby
 authors: 
 - Jason M. Adams
@@ -10,7 +10,7 @@ autorequire:
 bindir: bin
 cert_chain: []
 
-date: 2008-11-24 00:00:00 -08:00
+date: 2008-11-25 00:00:00 -08:00
 default_executable: 
 dependencies: []