Adding converter for graph dataset

deepform/data/graph_dataset.py

@@ -0,0 +1,41 @@
+import glob
+
+import pandas as pd
+from pathlib import Path
+
+from spektral.data import Dataset, Graph
+from spektral.data.loaders import DisjointLoader
+from deepform.data.add_features import read_adjacency
+from deepform.common import DATA_DIR, TOKEN_DIR, TRAINING_DIR, TRAINING_INDEX
+
+
+class GraphDataset(Dataset):
+    """
+    Graph Dataset for Spektral
+    """
+    def __init__(self, **kwargs):
+        super().__init__(**kwargs)
+
+    def read(self):
+        output = list()
+        for file in TRAINING_DIR.glob('*.parquet'):
+            try:
+                file_name = Path(file).stem
+                data = pd.read_parquet(file)
+                x = data.loc[:, data.columns != 'label'].to_numpy()
+                y = data.loc[:, data.columns == 'label'].to_numpy()
+                # if Path(f'{file_name}.graph.npz').exists():
+                adj = read_adjacency(f'data/training/416903-collect-files-39738-political-file-2012-non.graph').toarray()
+                output.append(Graph(x=x, a=adj, y=y))
+            except OSError:
+                pass
+        return output
+
+dataset = GraphDataset()
+
+# Create a DataLoader
+# dataloader = DisjointLoader(dataset, node_level=True, batch_size=8)
+
+# Iterate through DataLoader for training
+# for batch in dataloader:
+#     print(batch)