Edited README.md some more.

README.md

@@ -159,7 +159,77 @@ _____________________________
 
 
 For convenience, and to make the syntax a bit more expressive, 
-some operators are overloaded. For instance 
+some operators are overloaded. For instance, the sum operator for `GraphTuples` is overloaded. It is possible to compute residual connections
+easilly as follows:
+
+```python
+def eval_full(G, core_steps =CORE_STEPS):
+    ## The actual computation
+    gi_str.graph_tuple_eval(G) # happens in-place.
+
+    for ncore in range(core_steps):
+        G += gcore.graph_tuple_eval(G.copy())
+
+    emb, node_output = node_regressor(G.nodes) # a function operating only on the nodes.
+    G.nodes = emb 
+    gcore_out = gt_to_global(G) # A function containing node-to-global and edge-to-global aggregators.
+
+    return gcore_out, node_output 
+
+```
+
+Currently global blocks are not implemented.
+The reason for this is that it is easy to implement them in a few lines when needed.
+Here is an example of implementing a graph-to-global function with the respective aggregators:
+
+```python
+def make_graph_tuple_to_global(insize = GN_STATE, agg_type = 'mean', global_state_out = 3, type_ = "node", local_bnn = LOCALBNN):
+
+    # It would have been much cleaner if I supported this in the library... 
+    agg_fcn = make_keras_simple_agg(insize,agg_type) # from ibk_gnns import make_keras_simple_agg
+    agg_fcn = agg_fcn[1]
+
+    # Constructing the node+edge -> global function. 
+    xx = Input(shape = (insize,))
+
+    if local_bnn == False:
+        out = Dense(insize, 'relu')(xx)
+        out = Dense(insize, 'relu')(out)
+    else:
+        out = Dense(insize)(xx)
+        out = tfp.layers.DenseLocalReparameterization(insize,'relu')(out) 
+        out = Dense(insize,'relu')(out)
+
+    out = Dense(global_state_out, activation = GLOBAL_OUTPUT_ACTIVATION, use_bias = False)(out)
+
+    global_fcn = Model(inputs = xx, outputs = out)
+    bnnlosses = global_fcn.losses
+
+    def fcn_node_and_edge(gt):
+        graph_indices_nodes = []
+        for k_,k in enumerate(gt.n_nodes):
+            graph_indices_nodes.extend(np.ones(k).astype("int")*k_)
+
+        graph_indices_edges = []
+        for k_,k in enumerate(gt.n_edges):
+            graph_indices_edges.extend(np.ones(k).astype("int")*k_)
+
+        o1 = agg_fcn(gt.nodes,graph_indices_nodes, gt.n_graphs) # node_to_global aggregation
+        o2 = agg_fcn(gt.edges,graph_indices_edges, gt.n_graphs) # edge_to_global aggregation
+        return global_fcn(o1+o2) # either concat or add the aggregated information.
+
+    def fcn_node(gt):
+        graph_indices_nodes = []
+        for k_,k in enumerate(gt.n_nodes):
+            graph_indices_nodes.extend(np.ones(k).astype("int")*k_)
+
+        o1 = agg_fcn(gt.nodes,graph_indices_nodes, gt.n_graphs)
+        return global_fcn(o1)
+
+    fcn_dict = {'node': fcn_node,'node_and_edge' : fcn_node_and_edge}
+    return fcn_dict[type_], global_fcn, bnnlosses
+
+```