experiments

Author: newling

CMakeLists.txt

@@ -8,7 +8,9 @@ cmake_minimum_required(VERSION 3.0)
 
 #preferred compiler. TODO, make this optional
 # (see https://cmake.org/Wiki/CMake_FAQ#How_do_I_use_a_different_compiler.3F).
-SET (CMAKE_CXX_COMPILER "clang++")
+
+#SET (CMAKE_CXX_COMPILER "clang++")
+#SET (CMAKE_CXX_COMPILER "g++")
 
 project(zentas)
 

python/experiments/experiments.py

@@ -45,7 +45,7 @@ def go(X, K):
   C_init = X[indices_init]
 
 
-  withskl = False
+  withskl = True
   witheak = True
 
   if withskl == True:
@@ -61,15 +61,15 @@ def go(X, K):
     sys.path.append("/home/james/clustering/idiap/eakmeans/lib")
     import kmeans
     teak0 = time.time()
-    bla = kmeans.get_clustering(X, K, verbose = 1, init = indices_init, n_threads = 1)
+    bla = kmeans.get_clustering(X, K, verbose = 1, init = indices_init, n_threads = 1)#, algorithm = "syin-ns")
     teak1 = time.time()
 
 
 
-  z = pyzentas.pyzen(K = K, metric = 'l2', energy = 'quadratic', max_itok = 0.0, max_time = .0, max_rounds = 0, seed = npr.randint(1000), patient = True, nthreads = 1, init = indices_init, with_tests = False, capture_output = False)
+  z = pyzentas.pyzen(K = K, metric = 'l2', energy = 'quadratic', max_itok = 0.0, max_time = .0, max_rounds = 0, seed = npr.randint(1000), patient = True, nthreads = 1, init = indices_init, with_tests = True, capture_output = True)
   tzen0 = time.time()
   print X.shape
-  tangerine =  z.den(X, do_vdimap = False, do_refinement = True, rf_max_rounds = 10000000)#8, rf_alg = "yinyang")
+  tangerine =  z.den(X, do_vdimap = False, do_refinement = True, rf_max_rounds = 10000000, rf_alg = "exponion")
   tzen1 = time.time()
   print tangerine["output"].split("\n")[-2::]
 
@@ -83,9 +83,12 @@ def go(X, K):
 
 
 
-K = 500
+K = 50
 npr.seed(1000)
-X = npr.randn(50000, 8)
+#X = npr.randn(50000, 8)
+X = rna.get_rna()[0:5000, 2::]
+X += 0.01*npr.randn(X.shape[0], X.shape[1])
+
 go(X, K)
 
 def sklearn_elkan():

zentas/CMakeLists.txt

@@ -1,4 +1,4 @@
-#include(EnableCompilerWarnings)
+include(EnableCompilerWarnings)
 add_definitions(-std=c++11)
 
 include_directories("include")

zentas/include/zentas/baseclusterer.hpp

@@ -424,8 +424,10 @@ class Clusterer<LpMetric<typename TData::DataIn>, TData, TOpt>
   virtual void set_center_sample_distance(
     size_t k, size_t k1, size_t j1, double threshold, double& distance) override final
   {
+    
     metric.set_distance(
-      centers_data.at_for_metric(k), cluster_datas[k1].at_for_metric(j1), threshold, distance);
+    centers_data.at_for_metric(k), cluster_datas[k1].at_for_metric(j1), threshold, distance);
+            
   }
 
   virtual void set_rf_center_sample_distance(

zentas/include/zentas/lpmetric.hpp

@@ -316,6 +316,7 @@ class TLpDistance : public BaseLpDistance<TNumber>
                     double                threshold,
                     double&               a_distance) override final
   {
+    
     ++ncalcs;
     a_distance = 0;
     TNumber diff;
@@ -333,7 +334,12 @@ class TLpDistance : public BaseLpDistance<TNumber>
       }
     }
     calccosts += dimension;
+    
+    
+    
+    //checked, not very slow.
     correct_distance(a_distance);
+    
   }
 
   void set_distance(const SparseVectorSample<TNumber>& a,
@@ -613,7 +619,10 @@ class LpMetric
      * about 2% when dimension = 2. 2%
      * slowdown is negligible, I'm going
      * for this clean code.  */
-    uptr_lpdistance->set_distance(a, b, threshold, distance);
+     
+
+      uptr_lpdistance->set_distance(a, b, threshold, distance);
+    
   }
 
   size_t get_ncalcs() const { return uptr_lpdistance->ncalcs; }