How to visualize provenance#

Note

This tutorial can be downloaded and run as a Jupyter Notebook: visualising_graphs.ipynb

The provenance graph of a database can be visually inspected, via graphviz, using both the python API and command-line interface.

See also

verdi graph generate -h

We first load a profile, containing the provenance graph (in this case we load an archive as the profile).

from aiida import load_profile
from aiida.common import LinkType
from aiida.orm import LinkPair
from aiida.storage.sqlite_zip import SqliteZipBackend
from aiida.tools.visualization import Graph, pstate_node_styles

profile = load_profile(SqliteZipBackend.create_profile('include/graph1.aiida'))
Warning: You are currently using a post release development version of AiiDA: 2.0.3.post0
Warning: Be aware that this is not recommended for production and is not officially supported.
Warning: Databases used with this version may not be compatible with future releases of AiiDA
Warning: as you might not be able to automatically migrate your data.

dict1_uuid = '0ea79a16-501f-408a-8c84-a2704a778e4b'
calc1_uuid = 'b23e692e-4e01-48dd-b515-4c63877d73a4'

The Graph class is used to store visual representations of the nodes and edges, which can be added separately or cumulatively by one of the graph traversal methods. The graphviz attribute returns a graphviz.Digraph instance, which will auto-magically render the graph in the notebook, or can be used to save the graph to file.

graph = Graph()
graph.add_node(dict1_uuid)
graph.add_node(calc1_uuid)
graph.graphviz
../_images/visualising_graphs_4_0.svg
graph.add_edge(
    dict1_uuid, calc1_uuid,
    link_pair=LinkPair(LinkType.INPUT_CALC, "input1"))
graph.graphviz
../_images/visualising_graphs_5_0.svg
graph.add_incoming(calc1_uuid)
graph.add_outgoing(calc1_uuid)
graph.graphviz
../_images/visualising_graphs_6_0.svg

The Graph can also be initialized with global style attributes, as outlined in the graphviz attributes table.

graph = Graph(node_id_type="uuid",
              global_node_style={"penwidth": 1},
              global_edge_style={"color": "blue"},
              graph_attr={"size": "8!,8!", "rankdir": "LR"})
graph.add_incoming(calc1_uuid)
graph.add_outgoing(calc1_uuid)
graph.graphviz
../_images/visualising_graphs_8_0.svg

Additionally functions can be parsed to the Graph initializer, to specify exactly how each node will be represented. For example, the pstate_node_styles() function colors process nodes by their process state.

def link_style(link_pair, **kwargs):
    return {"color": "blue"}

graph = Graph(node_style_fn=pstate_node_styles,
              link_style_fn=link_style,
              graph_attr={"size": "8!,8!", "rankdir": "LR"})
graph.add_incoming(calc1_uuid)
graph.add_outgoing(calc1_uuid)
graph.graphviz
../_images/visualising_graphs_10_0.svg

Edges can be annotated by one or both of their edge label and link type.

graph = Graph(graph_attr={"size": "8!,8!", "rankdir": "LR"})
graph.add_incoming(calc1_uuid,
                   annotate_links="both")
graph.add_outgoing(calc1_uuid,
                   annotate_links="both")
graph.graphviz
../_images/visualising_graphs_12_0.svg

The recurse_descendants() and recurse_ancestors() methods can be used to construct a full provenance graph.

graph = Graph(graph_attr={"size": "8!,8!", "rankdir": "LR"})
graph.recurse_descendants(
    dict1_uuid,
    origin_style=None,
    include_process_inputs=True,
    annotate_links="both"
)
graph.graphviz
../_images/visualising_graphs_14_0.svg

The link types can also be filtered, to view only the ‘data’ or ‘logical’ provenance.

graph = Graph(graph_attr={"size": "8,8!", "rankdir": "LR"})
graph.recurse_descendants(
    dict1_uuid,
    origin_style=None,
    include_process_inputs=True,
    annotate_links="both",
    link_types=("input_calc", "create")
)
graph.graphviz
../_images/visualising_graphs_16_0.svg
graph = Graph(graph_attr={"size": "8,8!", "rankdir": "LR"})
graph.recurse_descendants(
    dict1_uuid,
    origin_style=None,
    include_process_inputs=True,
    annotate_links="both",
    link_types=("input_work", "return")
)
graph.graphviz
../_images/visualising_graphs_17_0.svg

If you wish to highlight specific node classes, then the highlight_classes option can be used to only color specified nodes:

graph = Graph(graph_attr={"size": "20,20", "rankdir": "LR"})
graph.recurse_descendants(
    dict1_uuid,
    highlight_classes=['Dict']
)
graph.graphviz
../_images/visualising_graphs_19_0.svg