added functionality to inject knn and snn

src/snc/helpers/hparam_functions.py

@@ -13,7 +13,7 @@
 e.g., knn info, distance matrix...
 '''
 
-def get_euclidean_infos(raw, emb, dist_parameter, dist_function, length, k):
+def get_euclidean_infos(raw, emb, dist_parameter, dist_function, length, k, snn_knn_matrix):
     raw_dist_matrix = dm.dist_matrix_gpu(raw)
     emb_dist_matrix = dm.dist_matrix_gpu(emb)
 
@@ -35,7 +35,7 @@ def get_euclidean_infos(raw, emb, dist_parameter, dist_function, length, k):
         "emb_knn"         : emb_knn_info
     }
 
-def get_predefined_infos(raw, emb, dist_parameter, dist_function, length, k):
+def get_predefined_infos(raw, emb, dist_parameter, dist_function, length, k, snn_knn_matrix):
     raw_dist_matrix = np.zeros((length, length))
     emb_dist_matrix = np.zeros((length, length))
 
@@ -65,9 +65,9 @@ def get_predefined_infos(raw, emb, dist_parameter, dist_function, length, k):
 
 
 
-def get_snn_infos(raw, emb, dist_parameter, dist_function, length, k):
+def get_snn_infos(raw, emb, dist_parameter, dist_function, length, k, snn_knn_matrix):
 
-    infos = get_euclidean_infos(raw, emb, dist_parameter, dist_function, length, k)
+    infos = get_euclidean_infos(raw, emb, dist_parameter, dist_function, length, k, snn_knn_matrix)
 
     # Compute snn matrix
     raw_snn_matrix = sk.snn_gpu(infos["raw_knn"], length, k)
@@ -91,6 +91,24 @@ def get_snn_infos(raw, emb, dist_parameter, dist_function, length, k):
 
     return infos
 
+def get_inject_snn_infos(raw, emb, dist_parameter, dist_function, length, k, snn_knn_matrix):
+    infos = {}
+    infos["raw_knn"] = snn_knn_matrix["raw_knn"]
+    infos["emb_knn"] = snn_knn_matrix["emb_knn"]
+
+    infos["raw_snn_matrix"] = snn_knn_matrix["raw_snn"]
+    infos["emb_snn_matrix"] = snn_knn_matrix["emb_snn"]
+
+    raw_snn_max		= np.max(infos["raw_snn_matrix"])
+    emb_snn_max		= np.max(infos["emb_snn_matrix"])
+
+    infos["raw_snn_matrix"] /= raw_snn_max
+    infos["emb_snn_matrix"] /= emb_snn_max
+
+    infos["raw_dist_matrix"] = 1 / (infos["raw_snn_matrix"] + dist_parameter["alpha"])
+    infos["emb_dist_matrix"] = 1 / (infos["emb_snn_matrix"] + dist_parameter["alpha"])
+
+    return infos
 
 '''
 Helper functions to extract a cluster
@@ -200,7 +218,7 @@ def get_predefined_cluster_distance(cluster_a, cluster_b, raw, emb, infos, dist_
 INSTALLING Hyperparameter functions
 '''
 
-def install_hparam(dist_strategy, dist_parameter, dist_function, cluster_strategy, raw, emb):
+def install_hparam(dist_strategy, dist_parameter, dist_function, cluster_strategy, snn_knn_matrix, raw, emb):
     get_infos = None
     get_a_cluster = None
     get_clusterinng = None 
@@ -219,6 +237,10 @@ def install_hparam(dist_strategy, dist_parameter, dist_function, cluster_strateg
         get_infos = get_predefined_infos
         get_a_cluster = get_a_cluster_naive
         get_cluster_distance = get_predefined_cluster_distance
+    elif dist_strategy == "inject_snn":
+        get_infos = get_inject_snn_infos
+        get_a_cluster = get_a_cluster_snn
+        get_cluster_distance = get_snn_cluster_distance
     else:
         raise Exception("Wrong strategy choice!! check dist_strategy ('" + dist_strategy + "')")
 
@@ -238,30 +260,32 @@ def install_hparam(dist_strategy, dist_parameter, dist_function, cluster_strateg
 
     return HparamFunctions(
         raw, emb, dist_parameter, dist_function,
-        get_infos, get_a_cluster, get_clustering, get_cluster_distance
+        get_infos, get_a_cluster, get_clustering, get_cluster_distance, snn_knn_matrix
     )
 
 
 class HparamFunctions():
     '''
     Saving raw, emb info and setting parameter
     '''
-    def __init__(self, raw, emb, dist_parameter, dist_function, get_infos, get_a_cluster, get_clustering, get_cluster_distance):
+    def __init__(self, raw, emb, dist_parameter, dist_function, get_infos, get_a_cluster, get_clustering, get_cluster_distance, snn_knn_matrix):
         self.raw = raw
         self.emb = emb
         self.length = len(self.raw)
         self.dist_parameter = dist_parameter
         self.dist_function  = dist_function
         self.k = dist_parameter["k"]
+        self.snn_knn_matrix = snn_knn_matrix
 
         ## Inject functions
         self.get_infos = get_infos
         self.get_a_cluster = get_a_cluster
         self.get_clustering = get_clustering
         self.get_cluster_distance = get_cluster_distance
 
+
     def preprocessing(self):
-        self.infos = self.get_infos(self.raw, self.emb, self.dist_parameter, self.dist_function, self.length, self.k)
+        self.infos = self.get_infos(self.raw, self.emb, self.dist_parameter, self.dist_function, self.length, self.k, self.snn_knn_matrix)
         dissim_matrix = self.infos["raw_dist_matrix"] - self.infos["emb_dist_matrix"]
 
         dissim_max = np.max(dissim_matrix)

src/snc/snc.py

@@ -17,7 +17,8 @@ def __init__(
                      "alpha": 0.1, "k": "sqrt"
                  },        
                  dist_function=None,       # inject predefined distance function
-                 cluster_strategy="dbscan" # determines the way to consider clusters
+                 cluster_strategy="dbscan", # determines the way to consider clusters
+                 snn_knn_matrix=None,  # inject predefined similarity matrix (dist_strategy should be "inject_snn")
                 ):
         self.raw  = np.array(raw, dtype=np.float64)
         self.emb  = np.array(emb, dtype=np.float64)
@@ -32,6 +33,7 @@ def __init__(
         self.dist_parameter   = dist_parameter
         self.dist_function    = dist_function
         self.cluster_strategy = cluster_strategy
+        self.snn_knn_matrix   = snn_knn_matrix
 
         ## target score
         self.cohev_score = None
@@ -63,7 +65,7 @@ def fit(self, record_vis_info=False):
 
         self.cstrat = hp.install_hparam(
             self.dist_strategy, self.dist_parameter, self.dist_function,
-            self.cluster_strategy,
+            self.cluster_strategy, self.snn_knn_matrix,
             self.raw, self.emb
         )