Merge pull request #110 from geometric-intelligence/feat-mantra

Mantra dataset

configs/dataset/simplicial/mantra_betti_numbers.yaml

@@ -0,0 +1,40 @@
+loader:
+  _target_: topobenchmark.data.loaders.MantraSimplicialDatasetLoader
+  parameters: 
+    data_domain: simplicial
+    data_type: topological 
+    data_name: MANTRA
+    data_dir: ${paths.data_dir}/${dataset.loader.parameters.data_domain}/${dataset.loader.parameters.data_type}
+    manifold_dim: 3
+    version: "v0.0.5"
+    task_variable: "betti_numbers" # Options: ['name', 'genus', 'orientable'] To use 'torsion_coefficients', 'betti_numbers' fix multilabel multiclass issue
+    model_domain: ${model.model_domain}
+
+# Data definition
+parameters:
+  # In the case of higher-order datasets we have multiple feature dimentions 
+  num_features: [1,1,1]
+  #num_classes: 2 # Num classes depents on the task_variable
+
+  # Dataset parameters
+  # task: classification     # TODO: adapt pipeline to support multilabel classification
+  # loss_type: cross_entropy # TODO: adapt pipeline to support multilabel classification
+  # monitor_metric: accuracy # TODO: adapt pipeline to support multilabel classification
+  task_level: graph 
+  data_seed: 0
+
+#splits
+split_params:
+  learning_setting: inductive 
+  data_split_dir: ${paths.data_dir}/data_splits/${dataset.loader.parameters.data_name}
+  data_seed: 0
+  split_type: random #'k-fold' # either "k-fold" or "random" strategies
+  k: 10 # for "k-fold" Cross-Validation
+  train_prop: 0.5 # for "random" strategy splitting
+
+# Dataloader parameters
+dataloader_params:
+  batch_size: 5
+  num_workers: 0
+  pin_memory: False
+  persistent_workers: False

configs/dataset/simplicial/mantra_genus.yaml

@@ -0,0 +1,40 @@
+loader:
+  _target_: topobenchmark.data.loaders.MantraSimplicialDatasetLoader
+  parameters: 
+    data_domain: simplicial
+    data_type: topological 
+    data_name: MANTRA
+    data_dir: ${paths.data_dir}/${dataset.loader.parameters.data_domain}/${dataset.loader.parameters.data_type}
+    manifold_dim: 2
+    version: "v0.0.5"
+    task_variable: "genus" # Options: ['name', 'genus', 'orientable'] To use 'torsion_coefficients', 'betti_numbers' fix multilabel multiclass issue
+    model_domain: ${model.model_domain}
+
+# Data definition
+parameters:
+  # In the case of higher-order datasets we have multiple feature dimentions 
+  num_features: [1,1,1]
+  num_classes: 8 # Num classes depents on the task_variable
+
+  # Dataset parameters
+  task: classification
+  loss_type: cross_entropy
+  monitor_metric: accuracy
+  task_level: graph 
+  data_seed: 0
+
+#splits
+split_params:
+  learning_setting: inductive 
+  data_split_dir: ${paths.data_dir}/data_splits/${dataset.loader.parameters.data_name}
+  data_seed: 0
+  split_type: random #'k-fold' # either "k-fold" or "random" strategies
+  k: 10 # for "k-fold" Cross-Validation
+  train_prop: 0.5 # for "random" strategy splitting
+
+# Dataloader parameters
+dataloader_params:
+  batch_size: 5
+  num_workers: 0
+  pin_memory: False
+  persistent_workers: False

configs/dataset/simplicial/mantra_name.yaml

@@ -0,0 +1,40 @@
+loader:
+  _target_: topobenchmark.data.loaders.MantraSimplicialDatasetLoader
+  parameters: 
+    data_domain: simplicial
+    data_type: topological 
+    data_name: MANTRA
+    data_dir: ${paths.data_dir}/${dataset.loader.parameters.data_domain}/${dataset.loader.parameters.data_type}
+    manifold_dim: 2
+    version: "v0.0.5"
+    task_variable: "name" # Options: ['name', 'genus', 'orientable'] To use 'torsion_coefficients', 'betti_numbers' fix multilabel multiclass issue
+    model_domain: ${model.model_domain}
+
+# Data definition
+parameters:
+  # In the case of higher-order datasets we have multiple feature dimentions 
+  num_features: [1,1,1]
+  num_classes: 8 # Num classes depents on the task_variable
+
+  # Dataset parameters
+  task: classification
+  loss_type: cross_entropy
+  monitor_metric: accuracy
+  task_level: graph 
+  data_seed: 0
+
+#splits
+split_params:
+  learning_setting: inductive 
+  data_split_dir: ${paths.data_dir}/data_splits/${dataset.loader.parameters.data_name}
+  data_seed: 0
+  split_type: random #'k-fold' # either "k-fold" or "random" strategies
+  k: 10 # for "k-fold" Cross-Validation
+  train_prop: 0.5 # for "random" strategy splitting
+
+# Dataloader parameters
+dataloader_params:
+  batch_size: 5
+  num_workers: 0
+  pin_memory: False
+  persistent_workers: False

configs/dataset/simplicial/mantra_orientation.yaml

@@ -0,0 +1,40 @@
+loader:
+  _target_: topobenchmark.data.loaders.MantraSimplicialDatasetLoader
+  parameters: 
+    data_domain: simplicial
+    data_type: topological 
+    data_name: MANTRA
+    data_dir: ${paths.data_dir}/${dataset.loader.parameters.data_domain}/${dataset.loader.parameters.data_type}
+    manifold_dim: 2
+    version: "v0.0.5"
+    task_variable: "orientable" # Options: ['name', 'genus', 'orientable'] To use 'torsion_coefficients', 'betti_numbers' fix multilabel multiclass issue
+    model_domain: ${model.model_domain}
+
+# Data definition
+parameters:
+  # In the case of higher-order datasets we have multiple feature dimentions 
+  num_features: [1,1,1]
+  num_classes: 2 # Num classes depents on the task_variable
+
+  # Dataset parameters
+  task: classification
+  loss_type: cross_entropy
+  monitor_metric: accuracy
+  task_level: graph 
+  data_seed: 0
+
+#splits
+split_params:
+  learning_setting: inductive 
+  data_split_dir: ${paths.data_dir}/data_splits/${dataset.loader.parameters.data_name}
+  data_seed: 0
+  split_type: random #'k-fold' # either "k-fold" or "random" strategies
+  k: 10 # for "k-fold" Cross-Validation
+  train_prop: 0.5 # for "random" strategy splitting
+
+# Dataloader parameters
+dataloader_params:
+  batch_size: 5
+  num_workers: 0
+  pin_memory: False
+  persistent_workers: False

configs/evaluator/classification.yaml

@@ -4,4 +4,4 @@ task_level: ${dataset.parameters.task_level}
 num_classes: ${dataset.parameters.num_classes}
 
 # Metrics
-metrics:  [accuracy, precision, recall, auroc] #Available options: accuracy, auroc, precision, recall
+metrics:  [accuracy, precision, recall, auroc] # Available options: accuracy, auroc, precision, recall

configs/run.yaml

@@ -4,9 +4,9 @@
 # order of defaults determines the order in which configs override each other
 defaults:
   - _self_
-  - dataset: graph/ZINC
-  - model: cell/topotune
-  - transforms: ${get_default_transform:${dataset},${model}}  #tree #${get_default_transform:${dataset},${model}} #no_transform 
+  - dataset: simplicial/mantra_orientation
+  - model: simplicial/scn
+  - transforms: ${get_default_transform:${dataset},${model}} #no_transform
   - optimizer: default
   - loss: default
   - evaluator: default

configs/transforms/liftings/cell2hypergraph_default.yaml

@@ -0,0 +1,2 @@
+defaults:
+  - /transforms/liftings: null

configs/transforms/liftings/cell2simplicial_default.yaml

@@ -0,0 +1,2 @@
+defaults:
+  - /transforms/liftings: null

configs/transforms/liftings/hypergraph2cell_default.yaml

@@ -0,0 +1,2 @@
+defaults:
+  - /transforms/liftings: null

configs/transforms/liftings/hypergraph2simplicial_default.yaml

@@ -0,0 +1,2 @@
+defaults:
+  - /transforms/liftings: null

configs/transforms/liftings/simplicial2graph_default.yaml

@@ -0,0 +1,2 @@
+defaults:
+  - /transforms/liftings: null

test/data/load/test_datasetloaders.py

@@ -36,10 +36,15 @@ def _gather_config_files(self, base_dir: Path) -> List[str]:
         """
         config_files = []
         config_base_dir = base_dir / "configs/dataset"
-        exclude_datasets = {# Below the datasets that have some default transforms manually overriten with no_transform,
+        # Below the datasets that have some default transforms manually overriten with no_transform,
+        exclude_datasets = {"karate_club.yaml",
+                            # Below the datasets that have some default transforms with we manually overriten with no_transform,
                             # due to lack of default transform for domain2domain
                             "REDDIT-BINARY.yaml", "IMDB-MULTI.yaml", "IMDB-BINARY.yaml", #"ZINC.yaml"
                             }
+
+        # Below the datasets that takes quite some time to load and process                            
+        self.long_running_datasets = {"mantra_name.yaml", "mantra_orientation.yaml", "mantra_genus.yaml", "mantra_betti_numbers.yaml",}
 
 
         for dir_path in config_base_dir.iterdir():
@@ -75,12 +80,16 @@ def _load_dataset(self, data_domain: str, config_file: str) -> Tuple[Any, Dict]:
             parameters = hydra.compose(
                 config_name="run.yaml",
                 overrides=[f"dataset={data_domain}/{config_file}", f"model=graph/gat"], 
-                return_hydra_config=True
-
+                return_hydra_config=True, 
             )
             dataset_loader = hydra.utils.instantiate(parameters.dataset.loader)
             print(repr(dataset_loader))
-            return dataset_loader.load()
+
+            if config_file in self.long_running_datasets:
+                dataset, data_dir = dataset_loader.load(slice=100)
+            else:
+                dataset, data_dir = dataset_loader.load()
+            return dataset, data_dir
 
     def test_dataset_loading_states(self):
         """Test different states and scenarios during dataset loading."""

test/utils/test_config_resolvers.py

@@ -47,6 +47,7 @@ def test_get_default_transform(self):
         out = get_default_transform("graph/ZINC", "cell/can")
         assert out == "dataset_defaults/ZINC"
 
+
     def test_get_required_lifting(self):
         """Test get_required_lifting."""
         out = get_required_lifting("graph", "graph/gat")
@@ -106,6 +107,15 @@ def test_infer_in_channels(self):
         in_channels = infer_in_channels(cfg.dataset, cfg.transforms)
         assert in_channels == [1433,1433,1433]
 
+        cfg = hydra.compose(config_name="run.yaml", overrides=["model=graph/gcn", "dataset=simplicial/mantra_orientation"], return_hydra_config=True)
+        in_channels = infer_in_channels(cfg.dataset, cfg.transforms)
+        assert in_channels == [1]
+
+        cfg = hydra.compose(config_name="run.yaml", overrides=["model=simplicial/scn", "dataset=graph/cocitation_cora"], return_hydra_config=True)
+        in_channels = infer_in_channels(cfg.dataset, cfg.transforms)
+        assert in_channels == [1433,1433,1433]
+
+
 
     def test_infer_num_cell_dimensions(self):
         """Test infer_num_cell_dimensions."""