fix: add raw_normalize and add_self_loop

Author: galogm

.ci/install-dev.sh

@@ -1,8 +1,6 @@
 rm -rf .env
 
-# # install python 3.8.16 using pyenv:
-# pyenv install 3.8.16
-# pyenv local 3.8.16
+bash .ci/py.sh
 
 # create and activate virtual environment
 python3 -m venv .env

.ci/install.sh

@@ -16,12 +16,11 @@ source .env/bin/activate
 # python -m pip install -U pip
 
 # # torch cuda 11.3
-python -m pip install torch==1.12 torchvision torchaudio --index-url https://download.pytorch.org/whl/cu113
+python -m pip install "torch>=1.12,<2.2" torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
 
 # dgl cuda 11.3
 # add a source if prefered: -i https://pypi.tuna.tsinghua.edu.cn/simple/
-python -m pip install dgl==1.1.0 -f https://data.dgl.ai/wheels/cu113/repo.html
-python -m pip install dglgo -f https://data.dgl.ai/wheels-test/repo.html
+python3 -m pip install dgl==2.1 -f https://data.dgl.ai/wheels/cu121/repo.html
 
 # install requirements
 # add a source if prefered: -i https://pypi.tuna.tsinghua.edu.cn/simple/

.ci/py.sh

@@ -0,0 +1,27 @@
+# Check if python3 command exists
+flag=0
+if command -v python3 &>/dev/null; then
+    # Extracting the version number
+    python_version=$(python3 --version 2>&1 | awk '{print $2}')
+
+    # Splitting the version number into major and minor parts
+    IFS='.' read -r -a version_parts <<< "$python_version"
+
+    major_version="${version_parts[0]}"
+    minor_version="${version_parts[1]}"
+
+    if [ "$major_version" -eq 3 ] && [ "$minor_version" -ge 8 ]; then
+        echo "Python version $python_version is greater than or equal to 3.8."
+    else
+        echo "Python version $python_version is less than 3.8."
+        flag=1
+    fi
+else
+    echo "Python>=3.8 is not installed."
+    flag=1
+fi
+# # install python 3.8, i.e., using pyenv:
+# pyenv install 3.8
+if [ "$flag" -eq 1 ]; then
+    command -v pyenv >/dev/null 2>&1 && pyenv local 3.8 && { echo >&2 "Install python 3.8 using pyenv successfully.";} || { echo >&2 "Pyenv is not installed. Did not install python 3.8 using pyenv. Please install mannually and retry"; exit 1;}
+fi

graph_datasets/load_data.py

@@ -1,11 +1,14 @@
 """Load Graph Datasets
 """
+
 # pylint:disable=protected-access
 import ssl
 from typing import Tuple
+from typing import Union
 
 import dgl
 import torch
+import torch.nn.functional as F
 from torch_geometric.data import Data
 from torch_geometric.utils import from_dgl
 
@@ -33,9 +36,11 @@ def load_data(
     verbosity: int = 0,
     source: str = "pyg",
     return_type: str = "dgl",
+    raw_normalize: bool = True,
     rm_self_loop: bool = True,
+    add_self_loop: bool = False,
     to_simple: bool = True,
-) -> Tuple[dgl.DGLGraph, torch.Tensor, int] or Data:
+) -> Union[Tuple[dgl.DGLGraph, torch.Tensor, int], Data]:
     """Load graphs.
 
     Args:
@@ -47,14 +52,18 @@ def load_data(
         source (str, optional): Source for data loading. Defaults to "pyg".
         return_type (str, optional): Return type of the graphs within ["dgl", "pyg"]. \
             Defaults to "dgl".
+        raw_normalize (str, optional): Row normalize the feature matrix. Defaults to True.
         rm_self_loop (str, optional): Remove self loops. Defaults to True.
+        add_self_loop (str, optional): Add self loops no matter what rm_self_loop is. \
+            Defaults to True.
         to_simple (str, optional): Convert to a simple graph with no duplicate undirected edges.
 
     Raises:
         NotImplementedError: Dataset unknown.
 
     Returns:
-        Tuple[dgl.DGLGraph, torch.Tensor, int]: [graph, label, n_clusters]
+        Tuple[dgl.DGLGraph, torch.Tensor, int]: [graph, label, n_clusters] or \
+        torch_geometric.data.Data
 
     Example:
         .. code-block:: python
@@ -143,9 +152,12 @@ def load_data(
             f"https://galogm.github.io/graph_datasets_docs/rst/table.html"
         )
 
-    # remove self loop and turn graphs into undirected ones
+    if raw_normalize:
+        graph.ndata["feat"] = F.normalize(graph.ndata["feat"], dim=1)
     if rm_self_loop:
-        graph = dgl.remove_self_loop(graph)
+        graph = graph.remove_self_loop()
+    if add_self_loop:
+        graph = graph.remove_self_loop().add_self_loop()
     if to_simple:
         graph = dgl.to_bidirected(graph, copy_ndata=True)
 
@@ -155,9 +167,13 @@ def load_data(
     new_label = torch.tensor(list(map(lambda x: old2new[x.item()], label)))
     graph.ndata["label"] = new_label
 
+    name = f"{dataset_name}_{source}"
+    graph.name = name
+
     if verbosity:
         print_dataset_info(
-            dataset_name=f"{source.upper()} undirected {dataset_name}\nwithout self-loops",
+            dataset_name=
+            f"{source.upper()} undirected {dataset_name}\n  add_self_loop={add_self_loop} rm_self_loop={rm_self_loop}",
             n_nodes=graph.num_nodes(),
             n_edges=graph.num_edges(),
             n_feats=graph.ndata["feat"].shape[1],
@@ -168,7 +184,7 @@ def load_data(
         return graph, new_label, n_clusters
 
     data = from_dgl(graph)
-    data.name = dataset_name
+    data.name = name
     data.num_classes = n_clusters
     data.x = data.feat
     data.y = data.label

requirements.txt

@@ -1,3 +1,4 @@
+ogb
 gdown
 wget
 dgl>=1.1

tests/test.py

@@ -1,5 +1,6 @@
 """Test
 """
+
 # pylint:disable=duplicate-code
 from graph_datasets import load_data
 from graph_datasets.data_info import DATASETS
@@ -11,6 +12,10 @@ def main(_source, _dataset):
         directory="./data",
         source=_source,
         verbosity=3,
+        raw_normalize=True,
+        rm_self_loop=True,
+        add_self_loop=True,
+        to_simple=True,
     )
 
     # import argparse