sample large graphs

code/explainer/node_explainer.py

@@ -1,14 +1,16 @@
-import os 
+import os
+import random
+
 import networkx as nx
 import numpy as np
-import random
 import torch
 from captum.attr import IntegratedGradients, LayerGradCam, Saliency
 from gnn.model import GraphConv, GraphConvolution
 from torch.autograd import Variable
 from torch_geometric.data import Data
 from torch_geometric.nn import MessagePassing
 from torch_geometric.utils import to_networkx
+from utils.gen_utils import get_subgraph, sample_large_graph
 
 from explainer.gnnexplainer import GNNExplainer, TargetedGNNExplainer
 from explainer.gnnlrp import GNN_LRP
@@ -220,17 +222,19 @@ def explain_subgraphx_node(model, data, node_idx, x, edge_index, edge_weight, ta
 def explain_zorro_node(model, data, node_idx, x, edge_index, edge_weight, target, device, args, include_edges=None):
     zorro = Zorro(model, device, num_hops = args.num_gc_layers)
     print('explain node', zorro.explain_node(node_idx, x, edge_index))
-    selected_nodes, selected_features, executed_selection = zorro.explain_node(node_idx, x, edge_index)[0]
-    selected_nodes = torch.Tensor(selected_nodes.squeeze())
-    selected_features = torch.Tensor(selected_features.squeeze())
-    print("node_attrs", selected_nodes)
-    print("node_feature_mask", selected_features)
-    print("executed_selection", executed_selection)
-    node_attr = np.array(selected_nodes)
-    edge_mask = node_attr_to_edge(edge_index, node_attr)
-    edge_mask = edge_mask.cpu().detach().numpy()
-    node_feature_mask = node_feature_mask.cpu().detach().numpy()
-    return edge_mask, node_feature_mask
+    explanation = zorro.explain_node(node_idx, x, edge_index, tau=0.85, recursion_depth=3)
+    print('explanation', explanation)
+    #selected_nodes, selected_features, executed_selection = zorro.explain_node(node_idx, x, edge_index, tau=0.85, recursion_depth=4)
+    #selected_nodes = torch.Tensor(selected_nodes.squeeze())
+    #selected_features = torch.Tensor(selected_features.squeeze())
+    #print("node_attrs", selected_nodes)
+    #print("node_feature_mask", selected_features)
+    #print("executed_selection", executed_selection)
+    #node_attr = np.array(selected_nodes)
+    #edge_mask = node_attr_to_edge(edge_index, node_attr)
+    #edge_mask = edge_mask.cpu().detach().numpy()
+    #node_feature_mask = node_feature_mask.cpu().detach().numpy()
+    return #edge_mask, node_feature_mask
 
 
 
@@ -247,13 +251,13 @@ def explain_pgexplainer_node(model, data, node_idx, x, edge_index, edge_weight,
         state_dict = torch.load(pgexplainer_saving_path)
         pgexplainer.load_state_dict(state_dict)
     else:
+        data = sample_large_graph(data)
         pgexplainer.train_explanation_network(data)
         print("Save PGExplainer model...")
         torch.save(pgexplainer.state_dict(), pgexplainer_saving_path)
         state_dict = torch.load(pgexplainer_saving_path)
         pgexplainer.load_state_dict(state_dict)
 
-
     edge_mask = pgexplainer.explain_node(model, node_idx, x, edge_index)
     edge_mask = edge_mask.cpu().detach().numpy()
     return edge_mask, None

code/explainer/zorro.py

@@ -10,7 +10,7 @@
 
 class Zorro(torch.nn.Module):
 
-    def __init__(self, model, device, num_hops=None, log=True, greedy=True, record_process_time=False, add_noise=False, samples=10):
+    def __init__(self, model, device, num_hops=None, log=True, greedy=True, record_process_time=False, add_noise=False, samples=100):
         super(Zorro, self).__init__()
         self.model = model
         self.log = log
@@ -427,7 +427,7 @@ def recursively_get_minimal_sets(self, initial_distortion, tau, possible_nodes,
 
         return minimal_nodes_and_features_sets
 
-    def explain_node(self, node_idx, full_feature_matrix, edge_index, tau=0.15, recursion_depth=np.inf,
+    def explain_node(self, node_idx, full_feature_matrix, edge_index, tau=0.85, recursion_depth=np.inf,
                      save_initial_improve=False):
         r"""Learns and returns a node feature mask and an edge mask that play a
         crucial role to explain the prediction made by the GNN for node

code/main.py

@@ -25,7 +25,7 @@
 from gnn.eval import gnn_scores_gc, gnn_scores_nc, gnn_accuracy
 from gnn.model import GCN, GcnEncoderGraph, GcnEncoderNode
 from gnn.train import train_graph_classification, train_node_classification, train_real
-from utils.gen_utils import gen_dataloader, get_labels, get_test_graphs, get_test_nodes
+from utils.gen_utils import gen_dataloader, get_labels, get_test_graphs, get_test_nodes, sample_large_graph
 from utils.graph_utils import get_edge_index_batch, split_batch
 from utils.io_utils import check_dir, create_data_filename, create_mask_filename, create_model_filename, load_ckpt, save_checkpoint
 from utils.parser_utils import arg_parse, get_data_args, get_graph_size_args
@@ -116,7 +116,6 @@ def main_real(args):
     ### Explainer ###
     list_test_nodes = get_test_nodes(data, model, args)
     mask_filename = create_mask_filename(args)
-
     edge_masks, node_feat_masks, Time = compute_edge_masks_nc(list_test_nodes, model, data, device, args)
     """
     if args.dataset.startswith("ebay"):

code/utils/gen_utils.py

@@ -17,6 +17,16 @@ def list_to_dict(preds):
         preds_dict[key] = np.array(preds_dict[key])
     return preds_dict
 
+def sample_large_graph(data):
+    if data.num_edges > 50000:
+        print("Too many edges, sampling large graph...")
+        node_idx = random.randint(0, data.num_nodes - 1)
+        x, edge_index, mapping, edge_mask, subset, kwargs = get_subgraph(node_idx, data.x, data.edge_index, num_hops=3)
+        data = data.subgraph(subset)
+        print(f'Sample size: {data.num_nodes} nodes and {data.num_edges} edges')
+    return data
+
+
 
 def get_subgraph(node_idx, x, edge_index, num_hops, **kwargs):
     num_nodes, num_edges = x.size(0), edge_index.size(1)

configs/real/config_real_5expe_testpred.json

@@ -25,19 +25,6 @@
     "dataset": ["cora", "facebook", "cornell", "actor", "chameleon"],
     "strategy": "topk",
     "params_list": "1,2,5,10,15,20,100",
-    "explainer_name": [
-      "random",
-      "distance",
-      "pagerank",
-      "sa",
-      "ig",
-      "gradcam",
-      "occlusion",
-      "basic_gnnexplainer",
-      "gnnexplainer",
-      "pgmexplainer",
-      "pgexplainer",
-      "subgraphx"
-    ]
+    "explainer_name": ["random", "pgexplainer"]
   }
 }

configs/real/topk/config_real_topk_test.json → configs/real/topk/config_real_topk_5expe_pg.json

@@ -10,19 +10,18 @@
   },
   "meta": {
     "group": "checkpoints/node_classification/real/topk",
-    "name": "test",
+    "name": "topk_real_5",
     "dest-arg": "yes",
     "dest-name": "dest"
   },
   "params": {
     "data_save_dir": "data",
-    "seed": 0,
-    "num_test": 1,
+    "seed": [0, 1, 2, 3, 4],
+    "num_test": 100,
     "explain_graph": "False",
-    "hard_mask": "False",
-    "true_label_as_target": "True",
+    "hard_mask": ["True", "False"],
+    "true_label_as_target": ["True", "False"],
     "strategy": "topk",
-    "directed": ["True", "False"],
     "params_list": "1,5,10,15,20,25,50,100",
     "dataset": [
       "cora",
@@ -36,19 +35,6 @@
       "chameleon",
       "squirrel"
     ],
-    "explainer_name": [
-      "random",
-      "distance",
-      "pagerank",
-      "sa",
-      "ig",
-      "gradcam",
-      "occlusion",
-      "basic_gnnexplainer",
-      "gnnexplainer",
-      "pgmexplainer",
-      "pgexplainer",
-      "subgraphx"
-    ]
+    "explainer_name": ["random", "pgexplainer"]
   }
 }

configs/real/topk/test.json

@@ -21,18 +21,7 @@
     "explain_graph": "False",
     "hard_mask": "True",
     "true_label_as_target": "True",
-    "dataset": [
-      "cora",
-      "pubmed",
-      "citeseer",
-      "facebook",
-      "cornell",
-      "texas",
-      "wisconsin",
-      "actor",
-      "chameleon",
-      "squirrel"
-    ],
+    "dataset": ["facebook", "squirrel"],
     "params_list": "1,5,10,15,20,25,50,100",
     "explainer_name": "pgexplainer"
   }

configs/syn/topk/config_top_edges.json

@@ -20,6 +20,18 @@
     "seed": [0, 1, 2, 3, 4],
     "top_acc": "True",
     "dataset": ["syn1", "syn3", "syn4", "syn5", "syn6"],
-    "explainer_name": ["random", "pgexplainer"]
+    "explainer_name": [
+      "random",
+      "distance",
+      "pagerank",
+      "sa",
+      "ig",
+      "occlusion",
+      "basic_gnnexplainer",
+      "gnnexplainer",
+      "pgmexplainer",
+      "pgexplainer",
+      "subgraphx"
+    ]
   }
 }