Add graph analyzation

notebooks/analyze_graphs.py

 # %%
 import pandas as pd
 import pickle
+import numpy as np
+import matplotlib as mpl
 import matplotlib.pyplot as plt
 import networkx as nx
 
@@ -15,43 +17,64 @@ argument_edges = pd.Series(
 fig, (ax1, ax2) = plt.subplots(1, 2, gridspec_kw={'width_ratios': [4, 1]}, figsize=(10, 5))
 fig.suptitle('Argument Edges')
 ax1.bar(argument_edges.index, argument_edges)
+ax1.set_xlabel('Topic number')
+ax1.set_ylabel('# Edges')
 ax2.boxplot(argument_edges)
 ax2.yaxis.set_ticklabels([])
+ax2.xaxis.set_ticklabels([])
+plt.savefig('argument_edges.png')
 
 # %%
 fig, (ax1, ax2) = plt.subplots(1, 2, gridspec_kw={'width_ratios': [4, 1]}, figsize=(10, 5))
 fig.suptitle('Argument Nodes')
 ax1.bar(argument_nodes.index, argument_nodes)
 ax2.boxplot(argument_nodes)
+ax1.set_xlabel('Topic number')
+ax1.set_ylabel('# Nodes')
 ax2.yaxis.set_ticklabels([])
+ax2.xaxis.set_ticklabels([])
+plt.savefig('argument_nodes.png')
 
 
 # %%
 
-fig, axis = plt.subplots(5, 10, figsize=(15, 10))
-for ax, elem in zip(axis.flatten(), graphs.items()):
+fig, axis = plt.subplots(1, 5, figsize=(13, 5))
+fig.suptitle('Argument Graphs per Topic')
+half_graphs = list(graphs.items())[:5]
+for ax, elem in zip(axis.flatten(), half_graphs):
     topic_nrb = elem[0]
-    g = elem[1]
-    f = nx.Graph()
-    fedges = filter(lambda x: g.degree()[x[0]] > 0 and g.degree()[x[1]] > 0, g.edges())
-    f.add_edges_from(fedges)
-    nx.draw(f, node_size=10, ax=ax)
+    ax.set_title(topic_nrb)
+    g = elem[1].to_undirected()
+    Gcc = g.subgraph(sorted(nx.connected_components(g), key=len, reverse=True)[0])
+    pos = nx.spring_layout(Gcc, seed=10396953)
+    nx.draw_networkx_nodes(Gcc, pos, ax=ax, node_size=20)
+    nx.draw_networkx_edges(Gcc, pos, ax=ax, alpha=0.4)
+plt.savefig('argument_graphs.png')
 
 # %%
-len(graphs)
-# %%
-ax
 
-# %%
-# %%
-
-fig, axis = plt.subplots(5, 10, figsize=(20, 10))
+fig, axis = plt.subplots(5, 10, figsize=(20, 15))
 for ax, elem in zip(axis.flatten(), graphs.items()):
     topic_nrb = elem[0]
+    ax.set_title(topic_nrb)
     g = elem[1]
     f = nx.Graph()
     fedges = filter(lambda x: g.degree()[x[0]] > 1 and g.degree()[x[1]] > 1, g.edges())
     f.add_edges_from(fedges)
     ax.hist(nx.pagerank(f).values())
 
+
+# %%
+degree_list = [g.degree() for g in graphs.values()]
+degree_list = [item[1] for sublist in degree_list for item in sublist]
+fig, ax = plt.subplots()
+ax.set_title('Degrees Histogramm of Argument Graphs (all Topics)')
+ax.set_yscale('log')
+ax.set_xlabel("Degree")
+ax.set_ylabel("# of Nodes")
+ax.hist(degree_list)
+
+plt.savefig('degree_histogramm.png')
+
+
 # %%

python/src/prototype/retrieval/argument_graph.py

@@ -35,6 +35,8 @@ class ArgumentGraph:
                     premise_argument_id = premise["_id"].split('_')[0]
                     if premise_argument_id not in graph:
                         graph.add_node(premise_argument_id)
+                    else:
+                        graph.nodes[premise_argument_id]['premises']
                     for conc_idx in range(len(premises_per_conclusions)):
                         sim = sim_matrices[conc_idx][prem_idx]
                         if sim >= self.threshold: