Merge pull request #28 from ChEB-AI/fix/new_pred_pipeline

New prediction pipeline

Author: sfluegel05

.gitignore

@@ -131,4 +131,9 @@ dmypy.json
 react-app/build/
 react-app/node_modules/
 
-.idea
+.idea
+
+data/
+backend/data/
+.vscode/
+backend/config.json

README.md

@@ -30,7 +30,6 @@ and change the path for each setting according to your setup.
 The ensemble can take any models that are implemented in [python-chebifier](https://github.com/ChEB-AI/python-chebifier). See the repository for example configurations. Common arguments for a model are:
  * `type`: one of the available [MODEL_TYPES](https://github.com/ChEB-AI/python-chebifier/blob/dev/chebifier/model_registry.py), e.g. `electra`,
  * `batch_size`: Number of molecules that are passed to the model at once,
- * `target_labels_path`: List of ChEBI classes (the `classes.txt` file that comes as part of a [ChEB-AI](https://github.com/ChEB-AI/python-chebai) dataset)
  * `classwise_weights_path` (optional): Weights that should be assigned to each class (i.e., trust scores calculated on a validation set with [this script](https://github.com/ChEB-AI/python-chebai/blob/dev/chebai/result/generate_class_properties.py)
 
 

backend/api/chemclass.py

@@ -169,7 +169,7 @@ class PredictionDetailApiHandler(Resource):
 
     def post(self):
         parser = reqparse.RequestParser()
-        parser.add_argument("type", type=str)
+        parser.add_argument("type", type=str, required=False, default="type")  # can be used to specify different types of requests in the future
         parser.add_argument("smiles", type=str)
         parser.add_argument("selectedModels", type=dict)
 

backend/config.template.json

@@ -6,7 +6,6 @@
             "type": "chebifier type (e.g. electra)",
             "ckpt_path": "path/to/checkpoint.ckpt",
             "batch_size": 32,
-            "target_labels_path": "path/to/classes.txt",
             "classwise_weights_path": "path/to/trust.json"
         }
     }