--- a
+++ b/app/__init__.py
@@ -0,0 +1,28 @@
+from omegaconf import OmegaConf
+
+from app import apis, core, datasets, models
+from app.core.utils import init_random
+
+
+def create_app(my_pipeline, device):
+    # Load dataset
+    dataset_cfg = OmegaConf.load(f"configs/_base_/datasets/{my_pipeline.dataset}.yaml")
+    # Merge config
+    cfg = OmegaConf.merge(dataset_cfg, my_pipeline)
+
+    print(cfg.model_type, cfg.model, cfg.task)
+    if cfg.model_type == "ml" and cfg.task == "los":
+        apis.ml_los_pipeline.start_pipeline(cfg)
+    elif cfg.model_type == "ml" and cfg.task == "outcome":
+        apis.ml_outcome_pipeline.start_pipeline(cfg)
+    elif cfg.model_type in ["dl", "ehr"] and cfg.task == "los":
+        apis.dl_los_pipeline.start_pipeline(cfg, device)
+    elif cfg.model_type in ["dl", "ehr"] and cfg.task == "outcome":
+        apis.dl_outcome_pipeline.start_pipeline(cfg, device)
+    elif cfg.model_type in ["dl", "ehr"] and cfg.task == "multitask":
+        apis.dl_multitask_pipeline.start_pipeline(cfg, device)
+        # apis.dl_multitask_pipeline.start_inference(cfg, device)
+    elif cfg.model_type in ["dl", "ehr"] and cfg.task == "twostage":
+        apis.dl_twostage_pipeline.start_pipeline(cfg, device)
+    elif cfg.model_type == "ml" and cfg.task == "twostage":
+        apis.ml_twostage_pipeline.start_pipeline(cfg)