updated demos

.pre-commit-config.yaml

@@ -15,7 +15,7 @@ repos:
       - id: check-added-large-files
         args:
           - --maxkb=2048
-      - id: trailing-whitespace
+#      - id: trailing-whitespace
       - id: requirements-txt-fixer
 
   # - repo: https://github.com/astral-sh/ruff-pre-commit

configs/dataset/transforms/feature_liftings/base_lifting.yaml

@@ -1,3 +1,3 @@
 _target_: topobenchmarkx.transforms.data_transform.DataTransform
-transform_name: "ConcatentionLifting"
+transform_name: "ProjectionSum"
 transform_type: null

configs/dataset/transforms/feature_liftings/concatenate.yaml

@@ -1,3 +1,3 @@
 _target_: topobenchmarkx.transforms.data_transform.DataTransform
-transform_name: "ProjectionLifting"
+transform_name: "ConcatentionLifting"
 transform_type: null

configs/dataset/us_county_demos.yaml

@@ -2,7 +2,7 @@ _target_: topobenchmarkx.io.load.loaders.GraphLoader
 
 defaults:
   - transforms: ${get_default_transform:graph,${model}}
-  
+
 
 # Data definition
 parameters:
@@ -18,6 +18,7 @@ parameters:
   num_classes: 1
   task: regression
   task_variable: 'Election' # options: ['Election', 'MedianIncome', 'MigraRate', 'BirthRate', 'DeathRate', 'BachelorRate', 'UnemploymentRate']
+  force_reload: True
   loss_type: mse
   monitor_metric: mse
   task_level: node

configs/model/cell/can.yaml

@@ -25,21 +25,12 @@ loss:
   task: ${dataset.parameters.task}
   loss_type: ${dataset.parameters.loss_type}
 
-# evaluator:
-#   _target_: topobenchmarkx.evaluators.evaluator.Evaluator
-#   metrics: [acc, rocauc]
-
 readout:
   _target_: topobenchmarkx.models.readouts.default_readouts.GNNBatchReadOut
   task_level: ${dataset.parameters.task_level}
   in_channels: ${parameter_multiplication:${model.backbone.out_channels},${model.backbone.heads}}
   out_channels: ${dataset.parameters.num_classes}
 
-# readout_workaround:
-#   _target_: topobenchmarkx.models.readout_workaround.ReadOutWorkaround
-#   backbone_outputs: ["x_0"]
-
-
 backbone_wrapper:
   _target_: topobenchmarkx.models.wrappers.default_wrapper.CANWrapper
   _partial_: true

configs/model/cell/cwn.yaml

@@ -12,7 +12,7 @@ backbone:
   in_channels_1: ${model.feature_encoder.out_channels}
   in_channels_2: ${model.feature_encoder.out_channels}
   hid_channels: ${model.feature_encoder.out_channels}
-  n_layers: 4
+  n_layers: 2
 
 loss:
   _target_: topobenchmarkx.models.losses.loss.DefaultLoss

configs/model/cell/cwn_dcm.yaml

@@ -9,7 +9,7 @@ feature_encoder:
 backbone:
   _target_: custom_models.cell.cwn_dcm.CWNDCM
   in_channels: ${model.feature_encoder.out_channels}
-  n_layers: 4
+  n_layers: 1
   dropout: 0.0
 
 loss:

configs/model/graph/gcn.yaml

@@ -9,7 +9,7 @@ backbone:
   _target_: torch_geometric.nn.models.GCN
   in_channels: ${model.feature_encoder.out_channels}
   hidden_channels: ${model.feature_encoder.out_channels}
-  num_layers: 3
+  num_layers: 1
   dropout: 0.0
   act: relu
 

configs/model/simplicial/sccn.yaml

@@ -3,13 +3,13 @@ _target_: topobenchmarkx.models.network_module.NetworkModule
 feature_encoder:
   _target_: topobenchmarkx.models.encoders.default_encoders.BaseFeatureEncoder
   in_channels: ${infer_in_channels:${dataset}} # ${dataset.parameters.num_features}
-  out_channels: 128
+  out_channels: 32
 
 backbone:
   _target_: topomodelx.nn.simplicial.sccn.SCCN
   channels: ${model.feature_encoder.out_channels}
   max_rank: 1
-  n_layers: 1
+  n_layers: 2
   update_func: "sigmoid"
 
 loss:

configs/model/simplicial/scn.yaml

@@ -14,7 +14,7 @@ backbone:
   in_channels_0: ${model.feature_encoder.out_channels}
   in_channels_1: ${model.feature_encoder.out_channels}
   in_channels_2: ${model.feature_encoder.out_channels}
-  n_layers: 1
+  n_layers: 3
 
 loss:
   _target_: topobenchmarkx.models.losses.loss.DefaultLoss

configs/train.yaml

@@ -4,8 +4,8 @@
 # order of defaults determines the order in which configs override each other
 defaults:
   - _self_
-  - dataset: ZINC
-  - model: cell/can #hypergraph/unignn2 #
+  - dataset: us_county_demos
+  - model: simplicial/scn #hypergraph/unignn2 #
   - evaluator: default
   - callbacks: default
   - logger: wandb # set logger here or use command line (e.g. `python train.py logger=tensorboard`)

topobenchmarkx/data/dataloader_fullbatch.py

@@ -89,9 +89,9 @@ def collate_fn(batch):
                         + running_idx[f"cell_running_idx_number_{cell_dim}"]
                     ).long()
 
-                    running_idx[f"cell_running_idx_number_{cell_dim}"] += (
-                        current_number_of_cells  # current_number_of_nodes
-                    )
+                    running_idx[
+                        f"cell_running_idx_number_{cell_dim}"
+                    ] += current_number_of_cells  # current_number_of_nodes
 
             elif x_key == "x_hyperedges":
                 cell_dim = x_key.split("_")[1]
@@ -112,9 +112,9 @@ def collate_fn(batch):
                         + running_idx[f"cell_running_idx_number_{cell_dim}"]
                     ).long()
 
-                    running_idx[f"cell_running_idx_number_{cell_dim}"] += (
-                        current_number_of_hyperedges
-                    )
+                    running_idx[
+                        f"cell_running_idx_number_{cell_dim}"
+                    ] += current_number_of_hyperedges
             else:
                 # Function Batch.from_data_list creates a running index automatically
                 pass

topobenchmarkx/data/us_county_demos_dataset.py

@@ -7,7 +7,7 @@
 from torch_geometric.data import Data, InMemoryDataset
 from torch_geometric.io import fs
 
-from topobenchmarkx.io.load.cornel_dataset import load_us_county_demos
+from topobenchmarkx.io.load.us_county_demos import load_us_county_demos
 from topobenchmarkx.io.load.download_utils import download_file_from_drive
 from topobenchmarkx.io.load.split_utils import random_splitting
 
@@ -92,6 +92,14 @@ def __init__(
         data.val_mask = torch.from_numpy(splits["valid"])
         data.test_mask = torch.from_numpy(splits["test"])
 
+        # Standardize the node features respecting train mask
+        data.x = (data.x - data.x[data.train_mask].mean(0)) / data.x[
+            data.train_mask
+        ].std(0)
+        data.y = (data.y - data.y[data.train_mask].mean(0)) / data.y[
+            data.train_mask
+        ].std(0)
+
         # Assign data object to self.data, to make it be prodessed by Dataset class
         self.data, self.slices = self.collate([data])
 
@@ -154,7 +162,9 @@ def process(self) -> None:
         Returns:
             None
         """
-        data = load_us_county_demos(self.raw_dir, year=self.parameters.year)
+        data = load_us_county_demos(
+            self.raw_dir, year=self.parameters.year, y_col=self.parameters.task_variable
+        )
 
         data = data if self.pre_transform is None else self.pre_transform(data)
         self.save([data], self.processed_paths[0])

topobenchmarkx/io/load/loaders.py

@@ -267,7 +267,10 @@ def load(self) -> CustomDataset:
             )
 
             if self.transforms_config is not None:
-                dataset = Preprocessor(data_dir, dataset, self.transforms_config)
+                # force_reload=True because in this datasets many variables can be trated as y
+                dataset = Preprocessor(
+                    data_dir, dataset, self.transforms_config, force_reload=True
+                )
 
             # We need to map original dataset into custom one to make batching work
             dataset = CustomDataset([dataset[0]])

topobenchmarkx/io/load/preprocessor.py

@@ -22,14 +22,22 @@ class Preprocessor(torch_geometric.data.InMemoryDataset):
         Additional arguments.
     """
 
-    def __init__(self, data_dir, data_list, transforms_config, **kwargs):
+    def __init__(
+        self, data_dir, data_list, transforms_config, force_reload=False, **kwargs
+    ):
         if isinstance(data_list, torch_geometric.data.Dataset):
             data_list = [data_list.get(idx) for idx in range(len(data_list))]
         elif isinstance(data_list, torch_geometric.data.Data):
             data_list = [data_list]
         self.data_list = data_list
         pre_transform = self.instantiate_pre_transform(data_dir, transforms_config)
-        super().__init__(self.processed_data_dir, None, pre_transform, **kwargs)
+        super().__init__(
+            self.processed_data_dir,
+            None,
+            pre_transform,
+            force_reload=force_reload,
+            **kwargs,
+        )
         self.save_transform_parameters()
         self.load(self.processed_paths[0])
 

topobenchmarkx/io/load/cornel_dataset.py → topobenchmarkx/io/load/us_county_demos.py

@@ -4,7 +4,7 @@
 import torch_geometric
 
 
-def load_us_county_demos(path, year=2012):
+def load_us_county_demos(path, year=2012, y_col="Election"):
     edges_df = pd.read_csv(f"{path}/county_graph.csv")
     stat = pd.read_csv(f"{path}/county_stats_{year}.csv", encoding="ISO-8859-1")
 
@@ -83,7 +83,7 @@ def load_us_county_demos(path, year=2012):
     stat = stat.drop(columns=["DEM", "GOP", "FIPS"])
 
     # Prediction col
-    y_col = "Election"  # TODO: Define through config file
+
     x_col = list(set(stat.columns).difference(set([y_col])))
 
     stat["MedianIncome"] = (

topobenchmarkx/models/wrappers/default_wrapper.py

@@ -114,14 +114,18 @@ def normalize_matrix(matrix):
         )
 
         # Propagate signal down
-        model_out["x_2"] = x_2
-        model_out["x_1"] = self.norm_1(
-            x_1 + self.agg_conv_1(model_out["x_2"], batch.incidence_2),
-            batch.batch_1,
-        )
-        model_out["x_0"] = self.norm_2(
-            x_0 + self.agg_conv_2(model_out["x_1"], batch.incidence_1), batch.batch
-        )
+        # model_out["x_2"] = x_2
+        # model_out["x_1"] = x_1
+        # model_out["x_0"] = x_0
+
+        # model_out["x_2"] = x_2
+        # model_out["x_1"] = self.norm_1(
+        #     x_1 + self.agg_conv_1(model_out["x_2"], batch.incidence_2),
+        #     batch.batch_1,
+        # )
+        # model_out["x_0"] = self.norm_2(
+        #     x_0 + self.agg_conv_2(model_out["x_1"], batch.incidence_1), batch.batch
+        # )
 
         return model_out
 

topobenchmarkx/transforms/data_transform.py

@@ -16,7 +16,7 @@
 )
 from topobenchmarkx.transforms.feature_liftings.feature_liftings import (
     ConcatentionLifting,
-    ProjectionLifting,
+    ProjectionSum,
     SetLifting,
 )
 from topobenchmarkx.transforms.liftings.graph2cell import CellCyclesLifting
@@ -30,7 +30,6 @@
 )
 
 TRANSFORMS = {
-    ###
     # Graph -> Hypergraph
     "HypergraphKHopLifting": HypergraphKHopLifting,
     "HypergraphKNearestNeighborsLifting": HypergraphKNearestNeighborsLifting,
@@ -40,7 +39,7 @@
     # Graph -> Cell Complex
     "CellCyclesLifting": CellCyclesLifting,
     # Feature Liftings
-    "ProjectionLifting": ProjectionLifting,
+    "ProjectionSum": ProjectionSum,
     "ConcatentionLifting": ConcatentionLifting,
     "SetLifting": SetLifting,
     # Data Manipulations

topobenchmarkx/transforms/feature_liftings/feature_liftings.py

@@ -2,7 +2,7 @@
 import torch_geometric
 
 
-class ProjectionLifting(torch_geometric.transforms.BaseTransform):
+class ProjectionSum(torch_geometric.transforms.BaseTransform):
     r"""Lifts r-cell features to r+1-cells by projection.
 
     Parameters
@@ -33,7 +33,7 @@ def lift_features(
             if f"x_{elem}" not in data:
                 idx_to_project = 0 if elem == "hyperedges" else int(elem) - 1
                 data["x_" + elem] = torch.matmul(
-                    data["incidence_" + elem].t(),
+                    abs(data["incidence_" + elem].t()),
                     data[f"x_{idx_to_project}"],
                 )
         return data

topobenchmarkx/transforms/liftings/graph_lifting.py

@@ -6,13 +6,13 @@
 
 from topobenchmarkx.transforms.feature_liftings.feature_liftings import (
     ConcatentionLifting,
-    ProjectionLifting,
+    ProjectionSum,
     SetLifting,
 )
 
 # Implemented Feature Liftings
 FEATURE_LIFTINGS = {
-    "projection": ProjectionLifting,
+    "projection": ProjectionSum,
     "concatenation": ConcatentionLifting,
     "set": SetLifting,
 }