download improvements, model analysis

training/birds/utils/kaggle.py

@@ -34,16 +34,18 @@ def download_competition_data(competition: str, input_path: str | Path) -> None:
         print(os.listdir(data_path))
 
 
-def download_dataset(owner: str, dataset: str, input_path: str | Path) -> None:
+def download_dataset(owner: str, dataset: str, input_path: str | Path) -> bool:
     """
     Downloads data from kaggle competition only if input folder is empty
     :param comptetition: string with the competition name id of kaggle
     :param input_path: path of the input folder
     """
+    downloaded = False
     data_path = Path(input_path)
     if not data_path.exists():
         data_path.mkdir(parents=True)
     if not any(data_path.iterdir()):
+        downloaded = True
         api.dataset_download_files(f"{owner}/{dataset}", path=data_path)
         with ZipFile(data_path / (dataset + ".zip"), "r") as zipObj:
             # Extract all the contents of zip file in current directory
@@ -51,3 +53,5 @@ def download_dataset(owner: str, dataset: str, input_path: str | Path) -> None:
         os.remove(data_path / (dataset + ".zip"))
 
         print(os.listdir(data_path))
+
+    return downloaded

training/environment.yml

@@ -10,7 +10,7 @@ dependencies:
   - pip
   - cudatoolkit=11.7
   - pytorch==1.13.1
-  - torchvision
+  - torchvision==0.14.1
   - pytorch-cuda=11.7
   - fastcore=1.5.28
   - fastai=2.7.11

training/models/.gitignore

@@ -0,0 +1,2 @@
+*
+!.gitignore