Setup for clean run

configs/downstream/all.yaml

@@ -1,65 +1,66 @@
 seed: 42
 data_dir: /scratch/SCRATCH_SAS/roman/Gothenburg/GIFFLAR/
 root_dir: /scratch/SCRATCH_SAS/roman/Gothenburg/GIFFLAR/data
+logs_dir: logs
 datasets:
-  - name: Immunogenicity
-    task: classification
-  - name: Glycosylation
-    task: classification
+  #- name: Immunogenicity
+  #  task: classification
+  #- name: Glycosylation
+  #  task: classification
   - name: Taxonomy_Domain
     task: multilabel
-  - name: Taxonomy_Phylum
-    task: multilabel
-  - name: Taxonomy_Class
-    task: multilabel
-  - name: Taxonomy_Order
-    task: multilabel
-  - name: Taxonomy_Family
-    task: multilabel
-  - name: Taxonomy_Genus
-    task: multilabel
-  - name: Taxonomy_Species
-    task: multilabel
+  #- name: Taxonomy_Phylum
+  #  task: multilabel
+  #- name: Taxonomy_Class
+  #  task: multilabel
+  #- name: Taxonomy_Order
+  #  task: multilabel
+  #- name: Taxonomy_Family
+  #  task: multilabel
+  #- name: Taxonomy_Genus
+  #  task: multilabel
+  #- name: Taxonomy_Species
+  #  task: multilabel
 pre-transforms:
 model:
-  - name: rf
-    n_estimators: 500
-    n_jobs: -1
-    random_state: 42
-  - name: svm
-    random_state: 42
-  - name: xgb
-    random_state: 42
+  #- name: rf
+  #   n_estimators: 500
+  #  n_jobs: -1
+  #  random_state: 42
+  #- name: svm
+  #  random_state: 42
+  #- name: xgb
+  #  random_state: 42
   - name: mlp
     hidden_dim: 1024
     batch_size: 256
     num_layers: 3
-    epochs: 100
+    epochs: 1
     patience: 30
     learning_rate: 0
     optimizer: Adam
   - name: sweetnet
-    hidden_dim: 768
-    batch_size: 256
-    epochs: 50
+    hidden_dim: 1024
+    batch_size: 512
+    epochs: 1
     patience: 30
     learning_rate: 0.001
     optimizer: Adam
-    suffix: _768
+    suffix:
   - name: gnngly
-    hidden_dim: 14
-    batch_size: 256
+    hidden_dim: 1024
+    batch_size: 512
     num_layers: 5
-    epochs: 200
+    epochs: 1
     patience: 30
-    learning_rate: 0
+    learning_rate: 0.001
     optimizer: Adam
-    suffix: _5_14
+    suffix:
   - name: gifflar
-    hidden_dim: 768
-    batch_size: 32
-    num_layers: 6
-    epochs: 100
+    hidden_dim: 1024
+    batch_size: 512
+    num_layers: 8
+    epochs: 1
     learning_rate: 0.001
     optimizer: Adam
-    suffix: _768_6
+    suffix:

configs/downstream/both.yaml

@@ -0,0 +1,39 @@
+seed: 42
+data_dir: /scratch/SCRATCH_SAS/roman/Gothenburg/GIFFLAR/
+root_dir: /scratch/SCRATCH_SAS/roman/Gothenburg/GIFFLAR/data
+logs_dir: logs
+datasets:
+  - name: Immunogenicity
+    task: classification
+  - name: Glycosylation
+    task: classification
+  - name: Taxonomy_Domain
+    task: multilabel
+  - name: Taxonomy_Kingdom
+    task: multilabel
+  - name: Taxonomy_Phylum
+    task: multilabel
+  - name: Taxonomy_Class
+    task: multilabel
+  - name: Taxonomy_Order
+    task: multilabel
+  - name: Taxonomy_Family
+    task: multilabel
+  - name: Taxonomy_Genus
+    task: multilabel
+  - name: Taxonomy_Species
+    task: multilabel
+pre-transforms:
+  LaplacianPE:
+    dim: 20
+  RandomWalkPE:
+    dim: 20
+model:
+  - name: gifflar
+    hidden_dim: 1024
+    batch_size: 512
+    num_layers: 8
+    epochs: 100
+    learning_rate: 0.001
+    optimizer: Adam
+    suffix: _both

configs/downstream/gnngly.yaml

@@ -1,35 +1,17 @@
 seed: 42
 data_dir: /scratch/SCRATCH_SAS/roman/Gothenburg/GIFFLAR/
 root_dir: /scratch/SCRATCH_SAS/roman/Gothenburg/GIFFLAR/data
+logs_dir: logs
 datasets:
-  # - name: Immunogenicity
-  #   task: classification
-  # - name: Glycosylation
-  #   task: classification
-  # - name: Taxonomy_Domain
-  #   task: multilabel
-  # - name: Taxonomy_Kingdom
-  #   task: multilabel
-  # - name: Taxonomy_Phylum
-  #   task: multilabel
-  # - name: Taxonomy_Class
-  #   task: multilabel
-  # - name: Taxonomy_Order
-  #   task: multilabel
-  # - name: Taxonomy_Family
-  #   task: multilabel
-  # - name: Taxonomy_Genus
-  #   task: multilabel
-  # - name: Taxonomy_Species
-  #   task: multilabel
+  - name: Taxonomy_Domain
+    task: multilabel
 pre-transforms:
 model:
   - name: gnngly
-    hidden_dim: 14
-    batch_size: 256
+    hidden_dim: 1024
+    batch_size: 512
     num_layers: 5
-    epochs: 200
+    epochs: 1
     patience: 30
-    learning_rate: 0
+    learning_rate: 0.001
     optimizer: Adam
-    suffix: _5_14

configs/downstream/dev.yaml → configs/downstream/lppe.yaml

@@ -1,7 +1,7 @@
 seed: 42
 data_dir: /scratch/SCRATCH_SAS/roman/Gothenburg/GIFFLAR/
 root_dir: /scratch/SCRATCH_SAS/roman/Gothenburg/GIFFLAR/data
-logs_dir: logs_1024_10
+logs_dir: logs
 datasets:
   - name: Immunogenicity
     task: classification
@@ -24,13 +24,14 @@ datasets:
   - name: Taxonomy_Species
     task: multilabel
 pre-transforms:
+  LaplacianPE:
+    dim: 20
 model:
   - name: gifflar
     hidden_dim: 1024
-    batch_size: 33
-    num_layers: 5
+    batch_size: 512
+    num_layers: 8
     epochs: 100
     learning_rate: 0.001
     optimizer: Adam
-    suffix: _1024_5
-
+    suffix: _lp

configs/downstream/rwpe.yaml

@@ -0,0 +1,37 @@
+seed: 42
+data_dir: /scratch/SCRATCH_SAS/roman/Gothenburg/GIFFLAR/
+root_dir: /scratch/SCRATCH_SAS/roman/Gothenburg/GIFFLAR/data
+logs_dir: logs
+datasets:
+  - name: Immunogenicity
+    task: classification
+  - name: Glycosylation
+    task: classification
+  - name: Taxonomy_Domain
+    task: multilabel
+  - name: Taxonomy_Kingdom
+    task: multilabel
+  - name: Taxonomy_Phylum
+    task: multilabel
+  - name: Taxonomy_Class
+    task: multilabel
+  - name: Taxonomy_Order
+    task: multilabel
+  - name: Taxonomy_Family
+    task: multilabel
+  - name: Taxonomy_Genus
+    task: multilabel
+  - name: Taxonomy_Species
+    task: multilabel
+pre-transforms:
+  RandomWalkPE:
+    dim: 20
+model:
+  - name: gifflar
+    hidden_dim: 1024
+    batch_size: 512
+    num_layers: 8
+    epochs: 100
+    learning_rate: 0.001
+    optimizer: Adam
+    suffix: _rw

gifflar/baselines/gnngly.py

@@ -31,8 +31,8 @@ def __init__(self, output_dim, task, **kwargs):
         """
         super().__init__(14, output_dim, task)
 
-        del self.convs
-        del self.head
+        # del self.convs
+        # del self.head
 
         # Five layers of plain graph convolution with a hidden dimension of 14.
         self.layers = [
@@ -71,7 +71,7 @@ def forward(self, batch):
             Dict holding the node embeddings, the graph embedding, and the final model prediction
         """
         # Extract atomic graph from the heterogeneous graph
-        x = batch["gnngly_x"]
+        x = batch["gnngly_x"].float()
         batch_ids = batch["gnngly_batch"]
         edge_index = batch["gnngly_edge_index"]
 

gifflar/train.py

@@ -36,7 +36,7 @@ def setup(**kwargs):
         pre_transform=get_pretransforms(**(kwargs["pre-transforms"] or {})), **data_config,
     )
     data_config["num_classes"] = datamodule.train.dataset_args["num_classes"]
-    logger = CSVLogger(kwargs["logs_dir"], name=kwargs["model"]["name"] + kwargs["model"].get("suffix", ""))
+    logger = CSVLogger(kwargs["logs_dir"], name=kwargs["model"]["name"] + (kwargs["model"].get("suffix", None) or ""))
     kwargs["dataset"]["filepath"] = str(data_config["filepath"])
     logger.log_hyperparams(kwargs)
     metrics = get_metrics(data_config["task"], data_config["num_classes"])