visual-graph-datasets 0.8.0

Installation

First clone this repository:

git clone https://github/username/visual_graph_datasets.git

Then install it like this:

cd visual_graph_datasets
pip3 install -e .

Command Line Interface

// Example for the dataset 'rb_dual_motifs'
python3 -m visual_graph_datasets.cli download "rb_dual_motifs"

python3 -m visual_graph_datasets.cli config

python3 -m visual_graph_datasets.cli list

Quickstart

The datasets are mainly intended to be used in combination with other packages, but this package provides some basic utilities to load and explore the datasets themselves within python programs.

import os
import typing as t
import matplotlib.pyplot as plt

from visual_graph_datasets.config import Config
from visual_graph_datasets.web import ensure_dataset
from visual_graph_datasets.data import load_visual_graph_dataset
from visual_graph_datasets.visualization.base import draw_image
from visual_graph_datasets.visualization.importances import plot_node_importances_border
from visual_graph_datasets.visualization.importances import plot_edge_importances_border

# This object will load the settings from the main config file. This config file contains options
# such as changing the default datasets folder and defining custom alternative file share providers
config = Config()
config.load()

# First of all we need to make sure that the dataset exists locally, this function will download it from
# the default file share provider if it does not exist.
ensure_dataset('rb_dual_motifs', config)

# Afterwards we can be sure that the datasets exists and can now load it from the default datasets path.
# The data will be loaded as a dictionary whose int keys are the indices of the corresponding elements
# and the values are dictionaries which contain all the relevant data about the dataset element,
# (Dataset format is explained below)
dataset_path = os.path.join(config.get_datasets_path(), 'rb_dual_motifs')
data_index_map: t.Dict[int, dict] = {}
_, data_index_map = load_visual_graph_dataset(dataset_path)

# Using this information we can visualize the ground truth importance explanation annotations for one
# element of the dataset like this.
index = 0
data = data_index_map[index]
# This is the dictionary which represents the graph structure of the dataset element. Descriptive
# string keys and numpy array values.
g = data['metadata']['graph']
fig, ax = plt.subplots(ncols=1, nrows=1, figsize=(10, 10))
draw_image(ax, image_path=data['image_path'])
plot_node_importances_border(
    ax=ax,
    g=g,
    node_positions=g['image_node_positions'],
    node_importances=g['node_importances_2'][:, 0],
)
plot_edge_importances_border(
    ax=ax,
    g=g,
    node_positions=g['image_node_positions'],
    edge_importances=g['edge_importances_2'][:, 0],
)
fig_path = os.path.join(os.getcwd(), 'importances.pdf')
fig.savefig(fig_path)

Dataset Format

Assuming the following shape definitions:

• V - the number of nodes in a graph

• E - the number of edges in a graph

• N - the number of node attributes / features associated with each node

• M - the number of edge attributes / features associated with each edge

• K - the number of importance channels

One such content dictionary which are the values of the two dicts returned by the function have the following nested dictionary structure:

• image_path: The absolute path to the image file that visualizes this element

• metadata_path: the absolute path to the metadata file

• metadata: A dict which contains all the metadata for that element

  • target: a 1d array containing the target values for the element. For classification this usually a one-hot encoded class vector already.

  • index: The canonical index of this element within the dataset

  • graph: A dictionary which contains the entire graph representation of this element.

    • node_indices: array of shape (V, 1) with the integer node indices.

    • node_attributes: array of shape (V, N)

    • edge_indices: array of shape (E, 2) which are the tuples of integer node indices that determine edges

    • edge_attributes: array of shape (E, M)

    • node_positions array of shape (V, 2) which are the xy positions of each node in pixel values within the corresponding image visualization of the element. This is the crucial information which is required to use the existing image representations to visualize attributional explanations!

    • (node_importances_{K}_{suffix}) array of shape (V, K) containing ground truth node importance explanations, which assign an importance value of 0 to 1 to each node of the graph across K channels. One dataset element may have none or multiple such annotations with different suffixes determining the number of explanation channels and origin.

    • (edge_importances_{K}_{suffix}) array of shape (E, K) containing ground truth edge importance explanations, which assign an importance value of 0 to 1 to each edge of the graph across K channels. One dataset element may have none or multiple such annotations with different suffixes determining the number of explanation channels and origin.

Datasets

Here is a list of the datasets currently included.

For more information about the individual datasets use the list command in the CLI (see above).

• TO BE DONE