rag.merge_hierarchical mean boundary fails with in_place_merge=False

[jni]

See the call to merge_hierarchical in this example:

labels2 = graph.merge_hierarchical(labels, g, thresh=0.08, rag_copy=False,
                                   in_place_merge=True,
                                   merge_func=merge_boundary,
                                   weight_func=weight_boundary)

If we change in_place_merge to False, we get the following traceback:

/Users/jni/conda/envs/ana3/lib/python3.5/site-packages/skimage/future/graph/graph_merge.py in merge_hierarchical(labels, rag, thresh, rag_copy, in_place_merge, merge_func, weight_func)
    127 
    128             merge_func(rag, src, dst)
--> 129             new_id = rag.merge_nodes(src, dst, weight_func)
    130             _revalidate_node_edges(rag, new_id, edge_heap)
    131 

<ipython-input-41-3e98b81231d5> in merge_nodes(self, src, dst, weight_func, in_place, extra_arguments, extra_keywords)
     13     for neighbor in neighbors:
     14         data = weight_func(self, src, new, neighbor, *extra_arguments,
---> 15                            **extra_keywords)
     16         self.add_edge(neighbor, new, attr_dict=data)
     17 

<ipython-input-69-dc9c8b5456f9> in weight_boundary(graph, src, dst, n)
      5 
      6     count_src = graph[src].get(n, default)['count']
----> 7     count_dst = graph[dst].get(n, default)['count']
      8 
      9     weight_src = graph[src].get(n, default)['weight']

KeyError: 'count'

> <ipython-input-69-dc9c8b5456f9>(7)weight_boundary()
      5 
      6     count_src = graph[src].get(n, default)['count']
----> 7     count_dst = graph[dst].get(n, default)['count']
      8 
      9     weight_src = graph[src].get(n, default)['weight']

I think it's because new edges are created with only a weight key, not a full complement of data attributes.

[jni]

CC @vighneshbirodkar

[DanielGutmann]

Same call to 'graph.merge_hierarchical' yet different error:

• Error occurs for any True/False combination of rag_copy= and in_place_merge=

TypeError                                 Traceback (most recent call last)
/Users/danielgutmann/anaconda/envs/med35/lib/python3.5/site-packages/skimage/future/graph/graph_merge.py in merge_hierarchical(labels, rag, thresh, rag_copy, in_place_merge, merge_func, weight_func)
    101         wt = data['weight']
    102         heap_item = [wt, n1, n2, True]
--> 103         heapq.heappush(edge_heap, heap_item)
    104 
    105         # Reference to the heap item in the graph

TypeError: unorderable types: float() < dict()`

[rfezzani]

graph.merge_hierarchical is no more in the API 😕. Closing

[jni]

@rfezzani I appreciate you trying to clean up many issues quickly, but except in very unambiguous cases, please give the original authors a chance to determine whether an issue can indeed be closed. In this case, the function is skimage.future.graph.merge_hierarchical and remains in the API, and this is still an issue that needs addressing.

[rfezzani]

Of course @jni and sorry if I am too quick closing issues 🙏 . It's a chance to revive some 😄. Maybe updating issue description may help 😉