spec & model prompt changes

rdagent/components/coder/data_science/model/eval_tests/model_execute.py

@@ -52,7 +52,7 @@
 print(execution_feedback_str)
 
 print("The second execution begins.\n")
-val_pred, test_pred, finalhypers = model_workflow(X=train_X, y=train_y, val_X=None, val_y=None, test_X=test_X, **hypers)
+val_pred, test_pred, finalhypers = model_workflow(X=train_X, y=train_y, val_X=None, val_y=None, test_X=test_X, hyper_params=hypers)
 execution_feedback_str = "The second Execution successful.\n"
 if val_pred is not None:
     execution_feedback_str += f"Validation predictions shape: {val_pred.shape}\n"

rdagent/components/coder/data_science/model/prompts.yaml

@@ -42,7 +42,7 @@ model_coder:
         =====Feedback:=====
         {{ former_failed_knowledge.feedback }}
         {% endfor %}
-        {% endif %}    
+        {% endif %}
 
     user: |-
         ---------Model Specification---------

rdagent/components/coder/data_science/raw_data_loader/prompts.yaml

@@ -13,6 +13,8 @@ spec:
     - Ensemble
     - The overall workflow
 
+    The specifications for each step should refer to the specifications in the previous sections and be tailored to the competition information provided.
+
     -----------Competition Information-----------
     {{ competition_info }}
 
@@ -123,31 +125,28 @@ spec:
     model: |-
       Model building specification text should adhere to the following requirements:
       1. Function Interface:
+        - Give a python function interface code with docstring.
         - The function name must be `model_workflow`.
-        - The function should include:
-          - Type annotations for all inputs and outputs.
-          - Input and output shapes:
-            - Input:
-              - `X`: A 4D NumPy array of shape `(num_samples, height, width, channels)`, where:
-                - `num_samples`: Number of training samples.
-                - `height` and `width`: Dimensions of the image (e.g., `224 x 224`).
-                - `channels`: Number of color channels (e.g., `3` for RGB).
-              - `y`: A 2D NumPy array of shape `(num_samples, 1)`, where `1` represents binary classification labels.
-              - Optional:
-                - `val_X`: Validation features of shape `(num_val_samples, height, width, channels)`.
-                - `val_y`: Validation labels of shape `(num_val_samples, 1)`.
-                - `test_X`: Test features of shape `(num_test_samples, height, width, channels)`.
-              - `hyper_params`: A dictionary of important hyperparameters for model configuration.
-            - Output:
-              - A tuple consisting of:
-                - `pred_val`: Predictions on validation data (`np.ndarray` of shape `(num_val_samples, 1)` or `None`).
-                - `pred_test`: Predictions on test data (`np.ndarray` of shape `(num_test_samples, 1)` or `None`).
-                - `hyper_params`: A dictionary of important hyperparameters for model configuration.
+        - Provide annotations for all inputs and outputs.
+        - Input:
+          - `X`: training features.
+          - `y`: training labels.
+          - Optional:
+            - `val_X`: Validation features.
+            - `val_y`: Validation labels.
+            - `test_X`: Test features.
+          - `hyper_params`: A dictionary of important hyperparameters for model configuration.
+        - Output:
+          - A tuple consisting of:
+            - `pred_val`: Predictions on validation data.
+            - `pred_test`: Predictions on test data.
+            - `hyper_params`: A dictionary of important hyperparameters for model configuration.
 
         - Include a clear and concise docstring to explain the function's purpose, its input parameters, and its expected return values.
 
       2. Precautions:
         - Ensure input arrays (`X`, `y`, `val_X`, `val_y`, `test_X`) have the correct shapes and consistent dimensions.
+        - You should check and handle outliers in your input data.
         - Use default values for hyperparameters if none are provided in `hyper_params`.
         - Return hyperparameters for retrain if not exists.
         - Perform model training on `X` and `y`, and evaluate using `val_X` and `val_y`.
@@ -276,6 +275,8 @@ data_loader_coder:
       You should follow the former code to improve it.
     {% endif %}
 
+    You should strictly follow the function interface specifications provided by the specification to implement the function.
+
 
 data_loader_eval:
   system: |-

rdagent/components/coder/data_science/workflow/__init__.py

@@ -50,7 +50,7 @@ def implement_one_task(
             load_data_code=workspace.file_dict["load_data.py"],
             feature_code=workspace.file_dict["feat01.py"],
             model_code=workspace.file_dict["model01.py"],
-            ensemble_code=workspace.file_dict["ens.py"],
+            ensemble_code=workspace.file_dict["ensemble.py"],
             latest_code=workspace.file_dict.get("main.py"),
             workflow_spec=workspace.file_dict["spec/workflow.md"],
         )