TICC fails with clusters with only one observation

.gitignore

@@ -1,2 +1,6 @@
 /**/*.pyc
 
+
+.idea/
+
+Results.txt

TICC_solver.py

@@ -16,7 +16,7 @@
 class TICC:
     def __init__(self, window_size=10, number_of_clusters=5, lambda_parameter=11e-2,
                  beta=400, maxIters=1000, threshold=2e-5, write_out_file=False,
-                 prefix_string="", num_proc=1, compute_BIC=False, cluster_reassignment=20):
+                 prefix_string="", num_proc=1, compute_BIC=False, cluster_reassignment=20, biased=False):
         """
         Parameters:
             - window_size: size of the sliding window
@@ -28,6 +28,7 @@ def __init__(self, window_size=10, number_of_clusters=5, lambda_parameter=11e-2,
             - write_out_file: (bool) if true, prefix_string is output file dir
             - prefix_string: output directory if necessary
             - cluster_reassignment: number of points to reassign to a 0 cluster
+            - biased: Using the biased or the unbiased covariance
         """
         self.window_size = window_size
         self.number_of_clusters = number_of_clusters
@@ -41,6 +42,7 @@ def __init__(self, window_size=10, number_of_clusters=5, lambda_parameter=11e-2,
         self.compute_BIC = compute_BIC
         self.cluster_reassignment = cluster_reassignment
         self.num_blocks = self.window_size + 1
+        self.biased = biased
         pd.set_option('display.max_columns', 500)
         np.set_printoptions(formatter={'float': lambda x: "{0:0.4f}".format(x)})
         np.random.seed(102)
@@ -328,7 +330,7 @@ def train_clusters(self, cluster_mean_info, cluster_mean_stacked_info, complete_
                 ##Fit a model - OPTIMIZATION
                 probSize = self.window_size * size_blocks
                 lamb = np.zeros((probSize, probSize)) + self.lambda_parameter
-                S = np.cov(np.transpose(D_train))
+                S = np.cov(np.transpose(D_train), bias=self.biased)
                 empirical_covariances[cluster] = S
 
                 rho = 1
@@ -398,4 +400,4 @@ def predict_clusters(self, test_data = None):
         # Update cluster points - using NEW smoothening
         clustered_points = updateClusters(lle_all_points_clusters, switch_penalty=self.switch_penalty)
 
-        return(clustered_points)
+        return(clustered_points)

UnitTest.py

@@ -64,6 +64,31 @@ def test_multiExample(self):
                 #Test failed
                 self.assertTrue(1==0)
 
+    def test_biased_vs_unbiased(self):
+        fname = "example_data.txt"
+        unbiased_ticc = TICC(window_size=1, number_of_clusters=8, lambda_parameter=11e-2, beta=600, maxIters=100,
+                             threshold=2e-5,
+                             write_out_file=False, prefix_string="output_folder/", num_proc=1)
+        (unbiased_cluster_assignment, unbiased_cluster_MRFs) = unbiased_ticc.fit(input_file=fname)
+
+        biased_ticc = TICC(window_size=1, number_of_clusters=8, lambda_parameter=11e-2, beta=600, maxIters=100,
+                           threshold=2e-5,
+                           write_out_file=False, prefix_string="output_folder/", num_proc=1, biased=True)
+        (biased_cluster_assignment, biased_cluster_MRFs) = biased_ticc.fit(input_file=fname)
+
+        np.testing.assert_array_equal(np.array(biased_cluster_assignment), np.array(unbiased_cluster_assignment), "Biased assignment is not equel to unbiased assignment!")
+
+    def test_failed_unbiased(self):
+        with self.assertRaises(Exception) as context:
+            # TICC will fail in Iteration 2, because cluster 9 has only one observation.
+            fname = "example_data.txt"
+            ticc = TICC(window_size=1, number_of_clusters=50, lambda_parameter=11e-2, beta=600, maxIters=100,
+                        threshold=2e-5,
+                        write_out_file=False, prefix_string="output_folder/", num_proc=1)
+            (cluster_assignment, cluster_MRFs) = ticc.fit(input_file=fname)
+
+        self.assertTrue('This is broken {}'.format(context.exception))
+
 
 if __name__ == '__main__':
     unittest.main()