log the actual image to wandb

app.py

@@ -78,7 +78,7 @@ if uploaded_file:
             print(f'Correct label for image {i}: {correct_label}')
 
             run.log({
-                'filepath': uploaded_file,
+                'image': wandb.Image(image),
                 'filetype': uploaded_file.type,
                 'predicted_label': id2label[prediction],
                 'predicted_label_id': prediction,

training.py

@@ -7,6 +7,7 @@ from constants import (PROCESSED_DATA_DIR,
                        EPOCHS,
                        BERT_BASE,
                        MAX_SEQUENCE_LENGHT,
+                       PROJECT_NAME,
                        FilePath,
                        PageMetadata,
                        ImageSize,
@@ -101,6 +102,21 @@ def prepare_data(
         test_frac=0.15,
     )
 
+    run = wandb.init(project_name=PROJECT_NAME, name='split-dataset')
+
+    split_dataset_artifact = wandb.Artifact('split-dataset-metadata', type='dataset')
+
+    train_table = wandb.Table(dataframe=train_df)
+    val_table = wandb.Table(dataframe=val_df)
+    test_table = wandb.Table(dataframe=test_df)
+
+    split_dataset_artifact.add(train_table, name='train_metadata')
+    split_dataset_artifact.add(val_table, name='val_metadata')
+    split_dataset_artifact.add(test_table, name='test_metadata')
+
+    run.log_artifact(split_dataset_artifact)
+    run.finish()
+
     print('Batching and shuffling the datasets')
     train_ds = dataset_from_dataframe(train_df)
     train_ds = prepare_dataset(train_ds, img_size, batch_size=BATCH_SIZE)
@@ -222,10 +238,6 @@ def train():
     )
 
 
-def evaluate():
-    return
-
-
 if __name__ = '__main__':
     train()
     evaluate()