From daf394657967209cfa7934707f4861def39ab040 Mon Sep 17 00:00:00 2001
From: John Hawkins <john.hawkins@john.hawkins.MacBook>
Date: Tue, 13 Apr 2021 21:34:17 +1000
Subject: [PATCH] Modifying test simulations

---
 minvime/estimator_classification.py |  8 ++---
 scripts/run_churn_simulation.py     | 54 +++++++++++++++++++++++++++++
 scripts/run_simulation_one.py       | 12 +++----
 3 files changed, 64 insertions(+), 10 deletions(-)
 create mode 100755 scripts/run_churn_simulation.py

diff --git a/minvime/estimator_classification.py b/minvime/estimator_classification.py
index cac503e..693a31f 100755
--- a/minvime/estimator_classification.py
+++ b/minvime/estimator_classification.py
@@ -73,9 +73,9 @@ def estimate_binary_model_requirements(tp, fp, tn, fn, cases, baserate, minroi=0
                 current_min_roi = roi
                 tprs = y
             combinations =  combinations + 1
-    print("Tested ", combinations, " different AUC plots")
-    print("Number of Exponents", len(beta_range))
-    print("Number of Alpha Weights", len(alpha_range))
+    #print("Tested ", combinations, " different AUC plots")
+    #print("Number of Exponents", len(beta_range))
+    #print("Number of Alpha Weights", len(alpha_range))
     return min_auc, min_precision, min_recall, np.array(fprates), tprs
 
 ######################################################################
@@ -91,7 +91,7 @@ def generate_roc_auc(fprates, alpha, beta):
 
 def calculate_peak_roi(fprates, tprates, tp, fp, tn, fn, num_pos, num_neg):
    """ Calculate the maximal ROI for a given ROC curve (defined by vectors of FPR and TPR) """
-   roi = -999999
+   roi = -99999999999999
    result_precision = 0.0
    result_recall = 0.0
    for index in range(len(fprates)):
diff --git a/scripts/run_churn_simulation.py b/scripts/run_churn_simulation.py
new file mode 100755
index 0000000..4084376
--- /dev/null
+++ b/scripts/run_churn_simulation.py
@@ -0,0 +1,54 @@
+"""
+Churn Simulation 
+
+We explore properties of a minimum viable model for churn problems with different criteria.
+
+In a churn situation every False Negative involves the cost of losing a customer.
+The cost of a False Positive is the cost of an uneccessary intervention.
+
+A True Positive means that you have a chance of mitigating the churn.
+Therefore the impact of a model making a True Positive is a reduction in the expected
+loss of a Churn. 
+
+For example, if churn involves a $10,000 loss per customer, and the
+probability of an intervention working is 20%. Then True Positive reduces the expected
+loss from $10,000 to $8,000. Or the model delivered $2,000 worth of value. 
+
+This script will produce a table of model performance criteria for problems of 
+varying definitions in terms of the costs of both churn and intervention.
+"""
+
+import sys
+import pandas as pd
+
+sys.path.append('../minvime')
+import estimator_classification as esti 
+
+fns = [-10000,-8000,-6000,-4000,-2000,-1000]
+fps = [-500,-400,-300,-200,-100]
+success_rate = 0.2
+
+tn = 0
+roi = 500000
+cases = 1000000
+baserate = 0.005
+
+rez = pd.DataFrame()
+
+for fn in fns:
+    for fp in fps:
+        # A true positive is a reduction in expected loss (after considering intervention cost)
+        tp = fn - (fn * success_rate) + fp
+        current_churn_loss = cases * baserate * fn
+        # To deliver ROI we need to lift ROI from the current churn losses.
+        minroi = current_churn_loss + roi
+        auc, prec, recall, fprs, tprs = esti.estimate_binary_model_requirements(
+            tp=tp, fp=fp, tn=tn, fn=fn, cases=cases, 
+            baserate=baserate, minroi=minroi
+        )
+        rez = rez.append({'Total Loss':current_churn_loss, 'Loss p Cust':fn, 'Intvn Cost':fp, 'auc':auc, 'precision':prec, 'recall':recall}, ignore_index=True)
+
+print("Churn Model Requirements")
+print("Criteria | Min ROI $%i | Cases %i | Base Churn Rate %f " % (roi,cases,baserate) )
+print(rez)
+
diff --git a/scripts/run_simulation_one.py b/scripts/run_simulation_one.py
index 06f8c98..f811883 100755
--- a/scripts/run_simulation_one.py
+++ b/scripts/run_simulation_one.py
@@ -4,8 +4,8 @@
 sys.path.append('../minvime')
 import estimator_classification as esti # The file ../minvime/estimator_classification.py
 
-tps = [20000,10000,8000,6000,4000,2000,1000,500]
-fps = [-1000,-900,-800,-600,-500,-400,-200,-100]
+tps = [20000,10000,8000,6000,4000,2000,1000]
+fps = [-900,-800,-600,-500,-400,-200,-100]
 
 tn = 0
 fn = 0
@@ -17,12 +17,12 @@
 
 for tp in tps:
     for fp in fps:
-        auc, prec, recall = esti.estimate_binary_model_requirements(
+        auc, prec, recall, fprs, tprs = esti.estimate_binary_model_requirements(
             tp=tp, fp=fp, tn=tn, fn=fn, cases=cases, 
             baserate=baserate, minroi=minroi
         )
-        print({'tp':tp, 'fp':fp, 'auc':auc, 'precision':prec, 'recall':recall})
-        rez = rez.append({'tp':tp, 'fp':fp, 'auc':auc, 'precision':prec, 'recall':recall}, ignore_index=True)
-
+        rez = rez.append({'TP Benefit':tp, 'FP Cost':fp, 'AUC':auc, 'precision':prec, 'recall':recall}, ignore_index=True)
+print("Model Requirements")
+print("Criteria | Min ROI $%i | Cases %i | Base Churn Rate %f " % (minroi,cases,baserate) )
 print(rez)