CONT-282: Calculated Markov chains can have rows with sums smaller th…

…an one.

clustinator/clustering.py

@@ -4,6 +4,7 @@
 
 from sklearn.cluster import DBSCAN
 import numpy as np
+import math
 
 # Data imports
 PATH = "../data/raw/"
@@ -45,7 +46,17 @@ def cluster_means(self):
         for label in np.unique(labels):
             chains_with_label = self.X[labels == label]
             mean_chain = sum(chains_with_label) / chains_with_label.shape[0]
-            # TODO: normalize rows to 1 (if a session misses an endpoint, that row will be 0 --> mean can be less than 1)
-            cluster_mean_dict[str(label)] = mean_chain.toarray()[0].tolist()
+            mean_chain = mean_chain.toarray()[0].tolist()
+            row_length = int(round(math.sqrt(len(mean_chain))))
+
+            for i in range(len(mean_chain) // row_length):
+                row_sum = sum(mean_chain[i * row_length:(i+1) * row_length])
+
+                if row_sum > 0 and abs(row_sum - 1.0) > 0.001:
+                    # if a session misses an endpoint, that row will be 0 --> mean can be less than 1
+                    for j in range(i * row_length, (i+1) * row_length):
+                        mean_chain[j] = mean_chain[j] / row_sum
+
+            cluster_mean_dict[str(label)] = mean_chain
 
         return cluster_mean_dict