Create app.py

app.py

@@ -0,0 +1,188 @@
+from fastapi import FastAPI, HTTPException
+from pydantic import BaseModel, Field, validator
+from typing import Optional
+import pandas as pd
+import joblib
+
+app = FastAPI()
+
+# Load models
+TFIDF_PATH = "models/tfidf_vectorizer.pkl"
+MODEL_PATH = "models/logreg_model.pkl"
+ENCODER_PATH = "models/label_encoders.pkl"
+
+tfidf_vectorizer = joblib.load(TFIDF_PATH)
+models = joblib.load(MODEL_PATH)
+label_encoders = joblib.load(ENCODER_PATH)
+
+# === Input schema ===
+class TransactionData(BaseModel):
+    Transaction_Id: str
+    Hit_Seq: int
+    Hit_Id_List: str
+    Origin: str
+    Designation: str
+    Keywords: str
+    Name: str
+    SWIFT_Tag: str
+    Currency: str
+    Entity: str
+    Message: str
+    City: str
+    Country: str
+    State: str
+    Hit_Type: str
+    Record_Matching_String: str
+    WatchList_Match_String: str
+    Payment_Sender_Name: Optional[str] = ""
+    Payment_Reciever_Name: Optional[str] = ""
+    Swift_Message_Type: str
+    Text_Sanction_Data: str
+    Matched_Sanctioned_Entity: str
+    Is_Match: int
+    Red_Flag_Reason: str
+    Risk_Level: str
+    Risk_Score: float
+    Risk_Score_Description: str
+    CDD_Level: str
+    PEP_Status: str
+    Value_Date: str
+    Last_Review_Date: str
+    Next_Review_Date: str
+    Sanction_Description: str
+    Checker_Notes: str
+    Sanction_Context: str
+    Maker_Action: str
+    Customer_ID: int
+    Customer_Type: str
+    Industry: str
+    Transaction_Date_Time: str
+    Transaction_Type: str
+    Transaction_Channel: str
+    Originating_Bank: str
+    Beneficiary_Bank: str
+    Geographic_Origin: str
+    Geographic_Destination: str
+    Match_Score: float
+    Match_Type: str
+    Sanctions_List_Version: str
+    Screening_Date_Time: str
+    Risk_Category: str
+    Risk_Drivers: str
+    Alert_Status: str
+    Investigation_Outcome: str
+    Case_Owner_Analyst: str
+    Escalation_Level: str
+    Escalation_Date: str
+    Regulatory_Reporting_Flags: bool
+    Audit_Trail_Timestamp: str
+    Source_Of_Funds: str
+    Purpose_Of_Transaction: str
+    Beneficial_Owner: str
+    Sanctions_Exposure_History: bool
+
+class PredictionRequest(BaseModel):
+    transaction_data: TransactionData
+
+@app.get("/")
+def root():
+    return {"status": "healthy", "message": "TF-IDF + LogReg API running"}
+
+@app.post("/validate")
+def validate_input(request: PredictionRequest):
+    return {"message": " Input is valid"}
+
+@app.post("/predict")
+def predict(request: PredictionRequest):
+    try:
+        input_df = pd.DataFrame([request.transaction_data.dict()])
+
+        # Create text_input for TF-IDF
+        text_input = f"""
+        Transaction ID: {input_df['Transaction_Id'].iloc[0]}
+        Origin: {input_df['Origin'].iloc[0]}
+        Designation: {input_df['Designation'].iloc[0]}
+        Keywords: {input_df['Keywords'].iloc[0]}
+        Name: {input_df['Name'].iloc[0]}
+        SWIFT Tag: {input_df['SWIFT_Tag'].iloc[0]}
+        Currency: {input_df['Currency'].iloc[0]}
+        Entity: {input_df['Entity'].iloc[0]}
+        Message: {input_df['Message'].iloc[0]}
+        City: {input_df['City'].iloc[0]}
+        Country: {input_df['Country'].iloc[0]}
+        State: {input_df['State'].iloc[0]}
+        Hit Type: {input_df['Hit_Type'].iloc[0]}
+        Record Matching String: {input_df['Record_Matching_String'].iloc[0]}
+        WatchList Match String: {input_df['WatchList_Match_String'].iloc[0]}
+        Payment Sender: {input_df['Payment_Sender_Name'].iloc[0]}
+        Payment Receiver: {input_df['Payment_Reciever_Name'].iloc[0]}
+        Swift Message Type: {input_df['Swift_Message_Type'].iloc[0]}
+        Text Sanction Data: {input_df['Text_Sanction_Data'].iloc[0]}
+        Matched Sanctioned Entity: {input_df['Matched_Sanctioned_Entity'].iloc[0]}
+        Red Flag Reason: {input_df['Red_Flag_Reason'].iloc[0]}
+        Risk Level: {input_df['Risk_Level'].iloc[0]}
+        Risk Score: {input_df['Risk_Score'].iloc[0]}
+        CDD Level: {input_df['CDD_Level'].iloc[0]}
+        PEP Status: {input_df['PEP_Status'].iloc[0]}
+        Sanction Description: {input_df['Sanction_Description'].iloc[0]}
+        Checker Notes: {input_df['Checker_Notes'].iloc[0]}
+        Sanction Context: {input_df['Sanction_Context'].iloc[0]}
+        Maker Action: {input_df['Maker_Action'].iloc[0]}
+        Customer Type: {input_df['Customer_Type'].iloc[0]}
+        Industry: {input_df['Industry'].iloc[0]}
+        Transaction Type: {input_df['Transaction_Type'].iloc[0]}
+        Transaction Channel: {input_df['Transaction_Channel'].iloc[0]}
+        Geographic Origin: {input_df['Geographic_Origin'].iloc[0]}
+        Geographic Destination: {input_df['Geographic_Destination'].iloc[0]}
+        Risk Category: {input_df['Risk_Category'].iloc[0]}
+        Risk Drivers: {input_df['Risk_Drivers'].iloc[0]}
+        Alert Status: {input_df['Alert_Status'].iloc[0]}
+        Investigation Outcome: {input_df['Investigation_Outcome'].iloc[0]}
+        Source of Funds: {input_df['Source_Of_Funds'].iloc[0]}
+        Purpose of Transaction: {input_df['Purpose_Of_Transaction'].iloc[0]}
+        Beneficial Owner: {input_df['Beneficial_Owner'].iloc[0]}
+        """
+
+        # Vectorize and predict
+        X_tfidf = tfidf_vectorizer.transform([text_input])
+        response = {}
+
+        for label, model in models.items():
+            proba = model.predict_proba(X_tfidf)[0]
+            pred_idx = proba.argmax()
+            decoded = label_encoders[label].inverse_transform([pred_idx])[0]
+            response[label] = {
+                "prediction": decoded,
+                "probabilities": {
+                    label_encoders[label].classes_[i]: float(p)
+                    for i, p in enumerate(proba)
+                }
+            }
+
+        return response
+
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=str(e))
+
+@app.get("/test")
+def test_model():
+    try:
+        sample_text = "Transaction flagged due to suspicious entity and origin. Needs compliance check."
+        X_tfidf = tfidf_vectorizer.transform([sample_text])
+        response = {}
+
+        for label, model in models.items():
+            proba = model.predict_proba(X_tfidf)[0]
+            pred_idx = proba.argmax()
+            decoded = label_encoders[label].inverse_transform([pred_idx])[0]
+            response[label] = {
+                "prediction": decoded,
+                "probabilities": {
+                    label_encoders[label].classes_[i]: float(p)
+                    for i, p in enumerate(proba)
+                }
+            }
+
+        return {"sample_input": sample_text, "predictions": response}
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=str(e))