Adding parallelizing script and a dummy config generator

explaining_framework/utils/config_gen.py

@@ -0,0 +1,74 @@
+import os
+
+from explaining_framework.utils.io import write_yaml
+from torch_geometric.data.makedirs import makedirs
+
+
+def divide_chunks(l, n):
+    for i in range(0, len(l), n):
+        yield l[i : i + n]
+
+
+if "__main__" == __name__:
+    config_folder = os.path.abspath(
+        os.path.join(os.path.dirname(__name__), "../../", "configs")
+    )
+    makedirs(config_folder)
+    explaining_folder = os.path.join(config_folder, "explaining")
+    makedirs(explaining_folder)
+    explainer_folder = os.path.join(config_folder, "explaining")
+    makedirs(explainer_folder)
+
+    DATASET = ["CIFAR10"]
+    EXPLAINER = [
+        "CAM",
+        "GradCAM",
+        "GNN_LRP",
+        "GradExplainer",
+        "GuidedBackPropagation",
+        "IntegratedGradients",
+        "PGExplainer",
+        "PGMExplainer",
+        "RandomExplainer",
+        "SubgraphX",
+        "GraphMASK",
+        "GNNExplainer",
+        "EIXGNN",
+        "SCGNN",
+    ]
+
+    for dataset_name in DATASET:
+        for chunk in divide_chunks(list(range(10000)), 500):
+            for model_kind in ["best", "worst"]:
+                for explainer_name in EXPLAINER:
+                    explaining_cfg = {}
+
+                    # explaining_cfg['adjust']['strategy']= 'rpns'
+                    # explaining_cfg['attack']['name']= 'all'
+                    explaining_cfg[
+                        "cfg_dest"
+                    ] = f"dataset={dataset_name}-model={model_kind}=explainer={explainer_name}-chunk=[{chunk[0]},{chunk[-1]}]"
+                    explaining_cfg["dataset"] = {}
+                    explaining_cfg["dataset"]["name"] = dataset_name
+                    explaining_cfg["dataset"]["item"] = chunk
+                    # explaining_cfg['explainer']['cfg']= 'default'
+                    explaining_cfg["explainer"] = {}
+                    explaining_cfg["explainer"]["name"] = explainer_name
+                    explaining_cfg["explanation_type"] = "phenomenon"
+                    # explaining_cfg['metrics']['accuracy']['name']='all'
+                    # explaining_cfg['metrics']['fidelity']['name']='all'
+                    # explaining_cfg['metrics']['sparsity']['name']='all'
+                    explaining_cfg["model"] = {}
+                    explaining_cfg["model"]["ckpt"] = model_kind
+                    explaining_cfg["model"][
+                        "path"
+                    ] = "/home/SIC/araison/test_ggym/pytorch_geometric/graphgym/results"
+                    # explaining_cfg['out_dir']='./explanation'
+                    # explaining_cfg['print']='both'
+                    # explaining_cfg['threshold']['config']['type']='all'
+                    # explaining_cfg['threshold']['value']['hard']=[0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9]
+                    # explaining_cfg['threshold']['value']['topk']=[2, 3, 5, 10, 20, 30, 50]
+                    write_yaml(
+                        explaining_cfg,
+                        explaining_folder + "/" + explaining_cfg["cfg_dest"] + ".yaml",
+                    )

parallel.sh

@@ -0,0 +1,22 @@
+CONFIG_DIR=$1
+MAX_JOBS=${3:-3}
+SLEEP=${4:-1}
+MAIN=${5:-main}
+GPU=${6:-0}
+
+(
+  trap 'kill 0' SIGINT
+  CUR_JOBS=0
+  for CONFIG in "$CONFIG_DIR"/*.yaml; do
+    if [ "$CONFIG" != "$CONFIG_DIR/*.yaml" ]; then
+      ((CUR_JOBS >= MAX_JOBS)) && wait -n
+      export CUDA_VISIBLE_DEVICES=$GPU
+      python3 $MAIN.py --explaining_cfg $CONFIG &
+      echo $CONFIG
+      sleep $SLEEP
+      ((++CUR_JOBS))
+    fi
+  done
+
+  wait
+)