Merge pull request #91 from kundajelab/hitscoring

v0.5.14.1 Nicer API for the density-adaptive hit scoring

modisco.egg-info/SOURCES.txt

@@ -9,6 +9,7 @@ modisco/core.py
 modisco/gammamix.py
 modisco/metaclusterers.py
 modisco/nearest_neighbors.py
+modisco/pattern_filterer.py
 modisco/util.py
 modisco/value_provider.py
 modisco.egg-info/PKG-INFO
@@ -39,6 +40,7 @@ modisco/cluster/phenograph/louvain/osx-hierarchy
 modisco/clusterinit/__init__.py
 modisco/clusterinit/memeinit.py
 modisco/hit_scoring/__init__.py
+modisco/hit_scoring/densityadapted_hitscoring.py
 modisco/hit_scoring/exemplar_based_hitscoring.py
 modisco/hit_scoring/fast_hit_scoring.py
 modisco/seqlet_embedding/__init__.py

modisco/cluster/core.py

@@ -186,6 +186,14 @@ def run_leiden(fileprefix, use_initclusters, n_vertices,
         if line.startswith("Membership:"):
             parse_membership = True 
 
+    if (parse_membership==False):
+        raise RuntimeError("----\nERROR:\n"
+                           +err.decode()
+                           +"\n----\nSTDOUT:\n"
+                           +out.decode()
+                           +"\n---\nARGS:\n"
+                           +str(args))
+
     return quality, np.array(membership)
 
 
@@ -234,10 +242,13 @@ def __call__(self, orig_affinity_mat, initclusters):
 
         #if an initclustering is specified, we would want to try the Leiden
         # both with and without that initialization and take the one that
-        # gets the best modularity
-        initclusters_to_try_list = [False]
-        if (initclusters is not None):
-            initclusters_to_try_list.append(True)
+        # gets the best modularity. EXCEPT when refinement is also specified.
+        if (self.refine):
+            initclusters_to_try_list = [True]
+        else:
+            initclusters_to_try_list = [False]
+            if (initclusters is not None):
+                initclusters_to_try_list.append(True)
 
 
         #write out the contents of affinity_mat and initclusters if applicable

modisco/cluster/run_leiden

@@ -14,8 +14,8 @@ def get_igraph(sources_idxs_file, targets_idxs_file, weights_file, n_vertices):
     weights = np.load(weights_file)
     g = ig.Graph(directed=None) 
     g.add_vertices(n_vertices) # this adds adjacency.shap[0] vertices
-    g.add_edges(list(zip(sources, targets))) 
-    g.es['weight'] = list(weights) 
+    g.add_edges(list(zip(sources.tolist(), targets.tolist()))) 
+    g.es['weight'] = weights.tolist()
     if g.vcount() != n_vertices: 
         print('WARNING: The constructed graph has only ' 
               +str(g.vcount())+' nodes. ' 
@@ -50,7 +50,8 @@ if __name__ == "__main__":
                           else np.load(args.initial_membership_file).tolist())
     seed = args.seed
 
-    weights = np.array(the_graph.es['weight']).astype(np.float64).tolist()
+    #weights = np.array(the_graph.es['weight']).astype(np.float64).tolist()
+    weights = (np.array(the_graph.es['weight']).astype(np.float64)).tolist()
 
     if (args.refine==False):
         partition = leidenalg.find_partition(
@@ -78,21 +79,29 @@ if __name__ == "__main__":
         #Github issue discussing things is here: https://github.com/vtraag/leidenalg/issues/61
         #move_nodes is what is used in the original louvain
         # (as per https://leidenalg.readthedocs.io/en/stable/advanced.html#optimiser)
-        #with move nodes
-        optimiser = leidenalg.Optimiser() 
-        optimiser.set_rng_seed(seed)
-        optimiser.move_nodes_constrained(refined_partition_movenodes,
-                                         constraining_partition)
+        #merge nodes is used for the refinement step of Leiden
+
+        #Due to segfault bug, I'm not doing move nodes:
+        # https://github.com/vtraag/leidenalg/issues/68
+
+        ##with move nodes
+        #optimiser = leidenalg.Optimiser() 
+        #optimiser.set_rng_seed(seed)
+        #optimiser.move_nodes_constrained(refined_partition_movenodes,
+        #                                 constraining_partition)
+
         #with merge nodes
         optimiser = leidenalg.Optimiser() 
         optimiser.set_rng_seed(seed)
         optimiser.merge_nodes_constrained(refined_partition_mergenodes,
                                           constraining_partition)
-        #take the partition with the best quality
-        partition = (refined_partition_movenodes if (
-                          refined_partition_movenodes.quality() >
-                          refined_partition_mergenodes.quality()) 
-                          else refined_partition_mergenodes)
+        partition = refined_partition_mergenodes
+
+        ##take the partition with the best quality
+        #partition = (refined_partition_movenodes if (
+        #                  refined_partition_movenodes.quality() >
+        #                  refined_partition_mergenodes.quality()) 
+        #                  else refined_partition_mergenodes)
 
     quality = partition.quality()
     print("########################")

modisco/coordproducers.py

@@ -630,7 +630,7 @@ def __call__(self, score_track, null_track, tnt_results=None):
                 val_transformer=precision_transformer) 
 
         else:
-            precision_transformer = tnt_results.precision_transformer
+            precision_transformer = tnt_results.val_transformer
             (precisiontransformed_score_track,
              precisiontransformed_bestwindowsizeidxs) =\
                 precision_transformer.transform_score_track(