modified: app/main.py

	modified:   app/utils.py
	modified:   setup.sh

app/main.py

@@ -224,6 +224,7 @@ async def predict_and_explain(
     surface: str | None = None,
     event: str | None = None,
     prob_min: float = 0.01,
+    top_k: int = 3,
 ):
     """
     Predict code APE.
@@ -247,7 +248,9 @@ async def predict_and_explain(
     text = [text_description]  # model needs a list of strings
     params = {"additional_var": [1] * len(text)}  # TBR
 
-    pred, confidence, all_scores = model.predict_and_explain(text, params)
+    pred, confidence, all_scores, all_scores_letters = model.predict_and_explain(
+        text, params, top_k=top_k
+    )
 
     response = process_response_explain(
         text=text,
@@ -257,6 +260,7 @@ async def predict_and_explain(
         all_scores=all_scores,
         prob_min=prob_min,
         libs=libs,
+        all_scores_letters=all_scores_letters,
     )
 
     return response

app/utils.py

@@ -312,6 +312,7 @@ def process_response_explain(
     all_scores: list[dict[str, float]],
     prob_min: float,
     libs: dict,
+    all_scores_letters,
 ):
     """
     Processes model predictions and generates response.
@@ -335,16 +336,20 @@ def process_response_explain(
             detail="The model is not confident enough to make a prediction (and explain it).",
         )
 
+    k = int(all_scores_letters.shape[0])  # top_k
     output_dict = {
-        str(1): {
-            "code": predictions[liasse_nb][-1].replace("__label__", ""),
-            "probabilite": float(confidence[liasse_nb][-1]),
-            "libelle": libs[predictions[liasse_nb][-1].replace("__label__", "")],
+        str(rank_pred + 1): {
+            "code": predictions[liasse_nb][-rank_pred - 1].replace("__label__", ""),
+            "probabilite": float(confidence[liasse_nb][-rank_pred - 1]),
+            "libelle": libs[predictions[liasse_nb][-rank_pred - 1].replace("__label__", "")],
+            "letter_attr": np.array(all_scores_letters).tolist()[
+                -rank_pred - 1
+            ],  # Converts numpy arrays to lists
+            "word_attr": np.array(all_scores[liasse_nb]).tolist()[-rank_pred - 1],
         }
+        for rank_pred in range(k)
     }
 
-    output_dict[text[liasse_nb]] = all_scores[liasse_nb]
-
     try:
         response = output_dict
         return response

setup.sh

@@ -8,7 +8,7 @@ pre-commit install
 export MLFLOW_S3_ENDPOINT_URL="https://$AWS_S3_ENDPOINT"
 export MLFLOW_TRACKING_URI=https://user-meilametayebjee-mlflow.user.lab.sspcloud.fr
 export MLFLOW_MODEL_NAME=fasttext-pytorch
-export MLFLOW_MODEL_VERSION=7
+export MLFLOW_MODEL_VERSION=8
 export API_USERNAME=username
 export API_PASSWORD=password
 export AUTH_API=False