class dgl.data.MinesweeperDataset(raw_dir=None, force_reload=False, verbose=True, transform=None)[source]

Bases: HeterophilousGraphDataset

Minesweeper dataset from the β€˜A Critical Look at the Evaluation of GNNs under Heterophily: Are We Really Making Progress? <https://arxiv.org/abs/2302.11640>’__ paper.

This dataset is inspired by the Minesweeper game. The graph is a regular 100x100 grid where each node (cell) is connected to eight neighboring nodes (with the exception of nodes at the edge of the grid, which have fewer neighbors). 20% of the nodes are randomly selected as mines. The task is to predict which nodes are mines. The node features are one-hot-encoded numbers of neighboring mines. However, for randomly selected 50% of the nodes, the features are unknown, which is indicated by a separate binary feature.


  • Nodes: 10000

  • Edges: 78804

  • Classes: 2

  • Node features: 7

  • 10 train/val/test splits

  • raw_dir (str, optional) – Raw file directory to store the processed data. Default: ~/.dgl/

  • force_reload (bool, optional) – Whether to re-download the data source. Default: False

  • verbose (bool, optional) – Whether to print progress information. Default: True

  • transform (callable, optional) – A transform that takes in a DGLGraph object and returns a transformed version. The DGLGraph object will be transformed before every access. Default: None


Number of node classes




>>> from dgl.data import MinesweeperDataset
>>> dataset = MinesweeperDataset()
>>> g = dataset[0]
>>> num_classes = dataset.num_classes
>>> # get node features
>>> feat = g.ndata["feat"]
>>> # get the first data split
>>> train_mask = g.ndata["train_mask"][:, 0]
>>> val_mask = g.ndata["val_mask"][:, 0]
>>> test_mask = g.ndata["test_mask"][:, 0]
>>> # get labels
>>> label = g.ndata['label']

Gets the data object at index.


The number of examples in the dataset.