changed timestep to date, made amounts smaller and more interpretable

app.py

@@ -5,7 +5,7 @@ from deeploy import Client
 from utils import get_request_body, get_fake_certainty, get_model_url, get_random_suspicious_transaction
 from utils import get_explainability_texts, get_explainability_values, send_evaluation, get_comment_explanation
 from utils import COL_NAMES, feature_texts
-from utils import create_data_input_table, create_table, ChangeButtonColour, get_weights
+from utils import create_data_input_table, create_table, ChangeButtonColour, get_weights, modify_datapoint
 
 logging.basicConfig(level=logging.INFO)
 
@@ -113,11 +113,12 @@ if st.session_state.got_explanation:
     prediction_log_id = st.session_state.prediction_log_id
     datapoint_pd = st.session_state.datapoint_pd
     certainty = st.session_state.certainty
+    datapoint = modify_datapoint(datapoint_pd)
 
     col1, col2 = st.columns(2)
 
     with col1:
-        create_data_input_table(datapoint_pd, COL_NAMES)
+        create_data_input_table(datapoint, COL_NAMES)
 
     with col2:
         st.subheader('AML Model Hit')
@@ -130,7 +131,7 @@ if st.session_state.got_explanation:
 
         explainability_texts, sorted_indices = get_explainability_texts(shap_values, feature_texts)
         weights = get_weights(shap_values, sorted_indices)
-        explainability_values = get_explainability_values(sorted_indices, datapoint_pd)
+        explainability_values = get_explainability_values(sorted_indices, datapoint)
         create_table(explainability_texts, explainability_values, weights, 'Important Suspicious Factors')
 
     st.subheader("")

utils.py

@@ -4,8 +4,11 @@ from random import randrange, uniform
 import pandas as pd
 import logging
 import numpy as np
+import random
+from datetime import datetime, timedelta
+from babel.numbers import format_currency
 
-COL_NAMES = ['Time step',
+COL_NAMES = ['Transaction date',
  'Transaction type',
  'Amount transferred',
  'Sender\'s initial balance',
@@ -20,7 +23,7 @@ COL_NAMES = ['Time step',
  'Sender ID',
  'Receiver ID']
 
-feature_texts = {0: "Time step: ", 1: "Amount transferred: ", 2: "Initial balance of sender: ", 3: "New balance of sender: ", 
+feature_texts = {0: "Date of transaction: ", 1: "Amount transferred: ", 2: "Initial balance of sender: ", 3: "New balance of sender: ", 
                  4: "Initial balance of recipient: ", 5: "New balance of recipient: ", 6: "Sender's balance was exactly credited: ",
                    7: "Receiver's balance was exactly credited: ", 8: "Transaction over 450.000: ", 9: "Frequent receiver of transactions: ", 10: "Receiver is merchant: ", 11: "Sender ID: ", 12: "Receiver ID: ",
                    13: "Transaction type is Cash out", 14: "Transaction type is Transfer", 15: "Transaction type is Payment", 16: "Transaction type is Cash in", 17: "Transaction type is Debit"}
@@ -59,8 +62,13 @@ def get_explainability_texts(shap_values, feature_texts):
     return positive_texts, sorted_positive_indices
 
 
-def get_explainability_values(pos_indices, datapoint):
-    data = datapoint.iloc[0].tolist()
+def random_past_date_from_last_year():
+    one_year_ago = datetime.now() - timedelta(days=365)
+    random_days = random.randint(0, (datetime.now() - one_year_ago).days)
+    random_date = one_year_ago + timedelta(days=random_days)
+    return random_date.strftime('%Y-%m-%d')
+
+def get_explainability_values(pos_indices, data):
     rounded_data = [round(value, 2) if isinstance(value, float) else value for value in data]
     transformed_data = transformation(input=rounded_data, categories=CATEGORIES)
     vals = []
@@ -72,11 +80,18 @@ def get_explainability_values(pos_indices, datapoint):
         vals.append(val)
     return vals
 
-# def get_weights(shap_values, sorted_indices):
-#     weights = [shap_values[x] for x in sorted_indices]
-#     total_sum = sum(weights)
-#     scaled_values = [val/total_sum for val in weights]
-#     return scaled_values
+def modify_datapoint(datapoint): # should return list, with correct numbers/amounts, and date
+    data = datapoint.iloc[0].tolist()
+    data[0] = random_past_date_from_last_year()
+    modified_amounts = data.copy()
+    if any(val > 12000 for val in data[2:7]):
+        modified_amounts[2:7] = [value / 100 if value != 0 else 0 for value in data[2:7]]
+    if any(val > 120000 for val in modified_amounts[2:7]):
+        new_list = [value / 10 if value != 0 else 0 for value in modified_amounts[2:7]]
+        modified_amounts[2:7] = new_list
+    rounded_data = [round(value, 2) if isinstance(value, float) else value for value in modified_amounts]
+    rounded_data[2:7] = [format_currency(value, 'EUR', locale='en_GB') for value in rounded_data[2:7]]
+    return rounded_data
 
 def get_weights(shap_values, sorted_indices, target_sum=0.95):
     weights = [shap_values[x] for x in sorted_indices]
@@ -141,9 +156,8 @@ def get_comment_explanation(certainty, explainability_texts, explainability_valu
     comment = f"Model certainty is {certainty}" + '\n''\n' + result
     return comment
 
-def create_data_input_table(datapoint, col_names):
+def create_data_input_table(data, col_names):
     st.subheader("Transaction details")
-    data = datapoint.iloc[0].tolist()
     data[7:12] = [bool(value) for value in data[7:12]]
     rounded_list = [round(value, 2) if isinstance(value, float) else value for value in data]
     df = pd.DataFrame({"Feature name": col_names, "Value": rounded_list })