Merge pull request #1073 from NASA-IMPACT/add_per_indicator_thresholding

add per indicator thrsholding and new dump

scripts/ej/cmr_to_models.py

@@ -69,7 +69,7 @@ def categorize_processing_level(level):
 # remove existing data
 EnvironmentalJusticeRow.objects.filter(destination_server=EnvironmentalJusticeRow.DestinationServerChoices.DEV).delete()
 
-ej_dump = json.load(open("backups/ej_dump_20240815_112916.json"))
+ej_dump = json.load(open("backups/ej_dump_20241017_133151.json.json"))
 for dataset in ej_dump:
     ej_row = EnvironmentalJusticeRow(
         destination_server=EnvironmentalJusticeRow.DestinationServerChoices.DEV,

scripts/ej/create_ej_dump.py

@@ -2,7 +2,7 @@
 inferences are supplied by the classification model. the contact point is Bishwas
 cmr is supplied by running
 github.com/NASA-IMPACT/llm-app-EJ-classifier/blob/develop/scripts/data_processing/download_cmr.py
-move to the serve like this: scp ej_dump_20240814_143036.json  sde:/home/ec2-user/sde_indexing_helper/backups/
+move to the server like this: scp ej_dump_20241017_133151.json  sde:/home/ec2-user/sde_indexing_helper/backups/
 """
 
 import json
@@ -19,20 +19,22 @@ def save_to_json(data: dict | list, file_path: str) -> None:
         json.dump(data, file, indent=2)
 
 
-def process_classifications(predictions: list[dict[str, float]], threshold: float = 0.5) -> list[str]:
+def process_classifications(predictions: list[dict[str, float]], thresholds: dict[str, float]) -> list[str]:
     """
-    Process the predictions and classify as follows:
-    1. If 'Not EJ' is the highest scoring prediction, return 'Not EJ' as the only classification
-    2. Filter classifications based on the threshold, excluding 'Not EJ'
-    3. Default to 'Not EJ' if no classifications meet the threshold
+    Process the predictions and classify based on the individual thresholds per indicator:
+    1. If 'Not EJ' is the highest scoring prediction, return 'Not EJ' as the only classification.
+    2. Filter classifications based on their individual thresholds, excluding 'Not EJ'.
+    3. Default to 'Not EJ' if no classifications meet the threshold.
     """
     highest_prediction = max(predictions, key=lambda x: x["score"])
 
     if highest_prediction["label"] == "Not EJ":
         return ["Not EJ"]
 
     classifications = [
-        pred["label"] for pred in predictions if pred["score"] >= threshold and pred["label"] != "Not EJ"
+        pred["label"]
+        for pred in predictions
+        if pred["score"] >= thresholds[pred["label"]] and pred["label"] != "Not EJ"
     ]
 
     return classifications if classifications else ["Not EJ"]
@@ -63,14 +65,14 @@ def remove_unauthorized_classifications(classifications: list[str]) -> list[str]
 def update_cmr_with_classifications(
     inferences: list[dict[str, dict]],
     cmr_dict: dict[str, dict[str, dict]],
-    threshold: float = 0.5,
+    thresholds: dict[str, float],
 ) -> list[dict[str, dict]]:
     """Update CMR data with valid classifications based on inferences."""
 
     predicted_cmr = []
 
     for inference in inferences:
-        classifications = process_classifications(predictions=inference["predictions"], threshold=threshold)
+        classifications = process_classifications(predictions=inference["predictions"], thresholds=thresholds)
         classifications = remove_unauthorized_classifications(classifications)
 
         if classifications:
@@ -84,17 +86,30 @@ def update_cmr_with_classifications(
 
 
 def main():
-    inferences = load_json_file("cmr-inference.json")
+    thresholds = {
+        "Not EJ": 0.80,
+        "Climate Change": 0.95,
+        "Disasters": 0.80,
+        "Extreme Heat": 0.50,
+        "Food Availability": 0.80,
+        "Health & Air Quality": 0.90,
+        "Human Dimensions": 0.80,
+        "Urban Flooding": 0.50,
+        "Water Availability": 0.80,
+    }
+
+    inferences = load_json_file("alpha-1.3-wise-vortex-42-predictions.json")
     cmr = load_json_file("cmr_collections_umm_20240807_142146.json")
 
     cmr_dict = create_cmr_dict(cmr)
 
-    predicted_cmr = update_cmr_with_classifications(inferences=inferences, cmr_dict=cmr_dict, threshold=0.8)
+    predicted_cmr = update_cmr_with_classifications(inferences=inferences, cmr_dict=cmr_dict, thresholds=thresholds)
 
     timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
     file_name = f"ej_dump_{timestamp}.json"
 
     save_to_json(predicted_cmr, file_name)
+    print(f"Saved to {file_name}")
 
 
 if __name__ == "__main__":