Recommended Python version is 3.7.6.
Packages are listed in requirements.txt.
- Install requirements:
pip install -r requirements- Run node2vec-a embedding:
Given a Networkx graph G, run node2vec-a to get embedding results:
from src.attr2vec.embedding import attributed_embeddingfrom
from src.util.graph_reading import init_attributed_graph, clean_attributed_graph
virtual_nodes = init_attributed_graph(G)
vectors, walks = attributed_embedding(
G, # Networkx graph
return_weight=return_weight, # 1/p
neighbor_weight=neighbor_weight, # 1/q
attribute_weight=attribute_weight, # 1/r
virtual_nodes=virtual_nodes # Extra virtual node list
)
clean_attributed_graph(G, vectors)Embedding vectors are stored in dict vectors.
- Run GEGraph layout:
from src.attr2vec.embedding import attributed_embeddingfrom src.from util.graph_reading import init_attributed_graph, clean_attributed_graph
pos = embedding_fr(
G, # Networkx graph
vectors=vectors, # Embedding vectors
wa=wa, # Weight of adjacent matrix
we=1-wa, # Weight of similarity matrix
cluster=node_class, # Node class dict, {"%node%": "%class%"}
tel=tel, # Intra-community truncation threshold
teh=teh # Inter-community truncation threshold
)Positions of node are stored in dict pos.
- Generate layout-preserving-aggregation view:
from src.util.key_words import graph_tf_idf
from src.aggregation.graph_aggregation import GraphAggregator
weights = graph_tf_idf(
G, # Networkx graph
cluster, # Clustering result or node class
walks, # Random walk paths. Used as context.
)
agg = GraphAggregator(
G, # Networkx graph
pos, # Node positions
group_attribute="group", # The community attribute of each node
fig_size=10, # View port width and height
ax_gap=0.05, # Gap between view port boundary and layout axis
is_curved=True, # Draw Bezier curves for edges linking two communities
attr_vectors=vectors, # Input embedding vectors
weights=weights
)
agg.generate_aggregations() # Precompute
agg.draw_aggregations( # Visualize
size_min=agg_size_min, # Minimum size of aggregation node
size_max=agg_size_max, # Maximum size of aggregation node
width_min=agg_width_min, # Minimum width of edges betweent aggregation nodes
width_max=agg_width_max, # Maximum width of edges betweent aggregation nodes
agg_alpha=agg_alpha, # Transparency of the background of clicked aggregation node
)
agg.set_events(fig)When the Matplotlib window shows, click any aggregation node to show the detail of the community.
- Generate related-nodes-searching view
from src.util.plot_drawing import draw_embedding_fr
draw_embedding_fr(
graph_file=graph_name, # Graph name
default_pos=pos, # Percomputed positions
attribute_weight=attribute_weight, # 1/r
neighbor_weight=neighbor_weight, # 1/q
return_weight=return_weight, # 1/p
seed=seed, # Seed for random
color=cluster, # Communities for node color
size_list=node_size, # Node sizes
width=edge_width, # Edge width
edge_alpha=edge_alpha, # Edge alpha
wa=wa, # Weight of adjacent matrix
we=we, # Weight of similarity matrix
tel=tel, # Intra-community truncation threshold
teh=teh, # Inter-community truncation threshold
fig=fig, # Matplotlib figure object
ax=ax, # Matplotlib axis object
ax_gap=0.05, # Gap between view port boundary and layout axis
fig_size=10, # View port width and height
click_on=True, # Enable node clicking
)Single click a node in the Matplotlib window, the similar nodes will be emphasized with different colors. Click the same node again, the nodes with attribute, local and global proximity will be shown recurrently.
- Run batch tests:
from tests.overall_test import run_overall_tests
run_overall_tests()Drawn graphs are stored in ./fig/<graphname>/<datetime>.
overall_graphs.json is a json configuration file for each graph in overall test：
{
"miserables": { // Graph name
"note": "overall", // Description of test
"seed": 6, // Seed for random
"k": 8, // K-Means: Cluster number
"d": 8, // node2vec-a: Dimension of vector
"walklen": 30, // node2vec-a: Length of random walk path
"attribute_weight": 0.6, // node2vec-a: 1/r
"neighbor_weight": 4, // node2vec-a: 1/q
"return_weight": 1, // node2vec-a: 1/p
"epochs": 15, // node2vec-a: Paths starting from each node
"tel": 0.4, // GEGraph: Truncation parameter for nodes in the same groups
"teh": 0.6, // GEGraph: Truncation parameter for nodes in different groups
"wa": 0.4, // GEGraph: Weight of adjacency matrix
"we": 0.6, // GEGraph: Weight of similarity matrix
"node_size": 30, // Drawing: Node size
"edge_width": 0.2, // Drawing: Edge width
"edge_alpha": 0.2, // Drawing: Edge alpha
"agg_size_min": 4000, // Application: The minimum size of the aggregated groups
"agg_size_max": 12000, // Application: The maximum size of the aggregated groups
"agg_width_min": 6, // Application: The minimum size of the links between the aggregated groups
"agg_width_max": 40, // Application: The maximum size of the links between the aggregated groups
"agg_alpha": 0.1, // Application: The alpha of the links between the aggregated groups
},
...
}