CoraLuDataset.py

import os
import numpy as np

from kgcnn.data.base import MemoryGraphDataset
from kgcnn.data.download import DownloadDataset


class CoraLuDataset(DownloadDataset, MemoryGraphDataset):
    r"""Store and process Cora dataset after `Lu et al. 2003 <https://www.aaai.org/Papers/ICML/2003/ICML03-066.pdf>`_ .

    Information in `Papers with code <https://paperswithcode.com/dataset/cora>`_ read:
    The Cora dataset consists of 2708 scientific publications classified into one of seven classes.
    The citation network consists of 5429 links. Each publication in the dataset is described by a 0/1-valued word
    vector indicating the absence/presence of the corresponding word from the dictionary.
    The dictionary consists of 1433 unique words.

    Downloaded from: `<https://linqs-data.soe.ucsc.edu/public/lbc/>`_ .

    References:

        (1) McCallum, A.K., Nigam, K., Rennie, J. et al. Automating the Construction of Internet Portals with Machine
            Learning. Information Retrieval 3, 127–163 (2000). https://doi.org/10.1023/A:1009953814988
        (2) Lu, Qing and Lise Getoor. “Link-Based Classification.” Encyclopedia of Machine Learning and Data
            Mining (2003).

    """

    download_info = {
        "dataset_name": "cora_lu",
        "data_directory_name": "cora_lu",
        "download_url": "https://linqs-data.soe.ucsc.edu/public/lbc/cora.tgz",
        # download_url = "https://linqs-data.soe.ucsc.edu/public/arxiv-mrdm05/arxiv.tar.gz"
        "download_file_name": 'cora.tgz',
        "unpack_tar": True,
        "unpack_zip": False,
        "unpack_directory_name": "cora_lu"
    }

    # Make cora graph that was published by Qing Lu, and Lise Getoor. "Link-based classification." ICML, 2003.
    # https://www.aaai.org/Papers/ICML/2003/ICML03-066.pdf

    def __init__(self, reload=False, verbose: int = 10):
        """Initialize Cora dataset after Lu et al. 2003.

        Args:
            reload (bool): Whether to reload the data and make new dataset. Default is False.
            verbose (int): Print progress or info for processing where 60=silent. Default is 10.
        """
        self.class_label_mapping = None
        # Use default base class init.

        MemoryGraphDataset.__init__(self, dataset_name="cora_lu", verbose=verbose)
        DownloadDataset.__init__(self, **self.download_info, reload=reload, verbose=verbose)

        if self.fits_in_memory:
            self.read_in_memory()

    def read_in_memory(self):
        """Load Cora data into memory and already split into items."""
        filepath = os.path.join(self.data_main_dir, self.data_directory_name, self.unpack_directory_name, "cora")

        ids = np.loadtxt(os.path.join(filepath, "cora.cites"))
        ids = np.array(ids, dtype="int64")
        open_file = open(os.path.join(filepath, "cora.content"), "r")
        lines = open_file.readlines()
        labels = [x.strip().split('\t')[-1] for x in lines]
        nodes = [x.strip().split('\t')[0:-1] for x in lines]
        nodes = np.array([[int(y) for y in x] for x in nodes], dtype="int64")
        open_file.close()
        # Match edge_indices not with ids but with edge_indices in nodes
        node_map = np.zeros(np.max(nodes[:, 0]) + 1, dtype="int64")
        idx_new = np.arange(len(nodes))
        node_map[nodes[:, 0]] = idx_new
        indexlist = node_map[ids]
        order1 = np.argsort(indexlist[:, 1], axis=0, kind='mergesort')  # stable!
        ind1 = indexlist[order1]
        order2 = np.argsort(ind1[:, 0], axis=0, kind='mergesort')
        indices = ind1[order2]
        # Class mappings
        class_label_mapping = {'Genetic_Algorithms': 0,
                               'Reinforcement_Learning': 1,
                               'Theory': 2,
                               'Rule_Learning': 3,
                               'Case_Based': 4,
                               'Probabilistic_Methods': 5,
                               'Neural_Networks': 6}
        self.class_label_mapping = class_label_mapping

        label_id = np.array([class_label_mapping[x] for x in labels], dtype="int")
        labels = np.expand_dims(label_id, axis=-1)
        labels = np.array(labels == np.arange(7), dtype="float")

        self.assign_property("node_attributes", [nodes[:, 1:]])
        self.assign_property("edge_indices", [indices])
        self.assign_property("edge_attributes", [np.ones_like(indices)[:, :1]])
        self.assign_property("node_labels", [labels])
        self.assign_property("node_number", [label_id])
        self.assign_property("edge_weights", [np.ones_like(indices)[:, :1]])
        # self.graph_adjacency = make_adjacency_from_edge_indices(indices, np.ones_like(indices)[:, 0])

        return self

# ds = CoraLuDataset()