updated compliance case, now success if >= 3.7

ml/notebooks/Complete_EDA.ipynb

@@ -379,7 +379,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.8.11"
+   "version": "3.8.8"
   },
   "orig_nbformat": 2
  },

ml/reports/Alarm model baseline.csv

-Method,c_harrell,c_uno,brier_score
-GradientBoosting,0.7192826813205445,0.8093132425965777,0.14487422609218523

ml/reports/Complete model baseline.csv

-clf,version,accuracy_mean,accuracy_std,precision_mean,precision_std,recall_mean,recall_std,roc_auc_mean,roc_auc_std,pr_auc_mean,pr_auc_std,f1_mean,f1_std
-KNN,Embedded,69.972,0.011,73.546,0.007,88.532,0.008,71.116,0.016,82.909,0.014,80.345,0.007
-SVM,Embedded,69.494,0.003,69.575,0.002,99.514,0.002,70.459,0.011,84.206,0.008,81.894,0.002
-LR,Embedded,71.152,0.002,71.582,0.003,96.84,0.007,72.068,0.01,85.501,0.007,82.314,0.001
-XGB,Embedded,71.264,0.003,72.266,0.002,95.016,0.003,70.929,0.014,84.333,0.01,82.093,0.002
-RF,Embedded,71.826,0.014,75.34,0.015,88.331,0.015,75.044,0.01,87.01,0.008,81.3,0.008
-MLP,Embedded,70.843,0.007,71.641,0.004,95.908,0.008,72.228,0.009,85.674,0.005,82.016,0.004

ml/reports/Compliance model baseline.csv

-clf,version,accuracy_mean,accuracy_std,precision_mean,precision_std,recall_mean,recall_std,roc_auc_mean,roc_auc_std,pr_auc_mean,pr_auc_std,f1_mean,f1_std
-KNN,Embedded,77.835,0.02,79.774,0.019,85.686,0.021,84.251,0.016,88.037,0.018,82.606,0.016
-SVM,Embedded,70.422,0.013,71.094,0.012,87.536,0.043,76.67,0.012,83.759,0.009,78.393,0.014
-LR,Embedded,71.393,0.019,72.721,0.018,85.69,0.044,79.617,0.023,86.508,0.018,78.594,0.018
-XGB,Embedded,70.989,0.01,72.605,0.018,85.028,0.037,77.139,0.02,84.093,0.021,78.246,0.009
-RF,Embedded,80.469,0.018,83.044,0.019,85.753,0.014,86.524,0.021,89.897,0.015,84.366,0.014
-MLP,Embedded,73.418,0.023,77.24,0.024,80.606,0.038,80.12,0.023,86.707,0.019,78.814,0.019

ml/src/analysis/evaluate_survival_case.py

@@ -23,7 +23,14 @@ def main():
                                      target_settings).load_data()
     X, y = dl.get_data()
 
-    model = GradientBoostingSurvivalAnalysis(n_estimators=200,
+    model = GradientBoostingSurvivalAnalysis(n_estimators=100,
+                                             learning_rate=0.1,
+                                             max_depth=14,
+                                             loss='coxph',
+                                             min_samples_split=4,
+                                             max_features='log2',
+                                             subsample=0.4,
+                                             dropout_rate=0.7,
                                              random_state=0)
 
     model.fit(X, y)

ml/src/model/train_compliance_model.py

@@ -21,14 +21,14 @@ def main():
 
     params = {"n_estimators": 400,
               "booster": "gbtree",
-              "max_depth": 5,
-              "gamma": 5,
-              "colsample_bytree": 1,
-              "min_child_weight": 8,
-              "reg_alpha": 10,
+              "max_depth": 8,
+              "gamma": 0,
+              "colsample_bytree": 0.3,
+              "min_child_weight": 9,
+              "reg_alpha": 18,
               "reg_lambda": 0.9,
               "learning_rate": 0.05,
-              "subsample": 0.8,
+              "subsample": 1,
               "use_label_encoder": False,
               "eval_metric": "logloss",
               "objective": "binary:logistic",

ml/src/tools/labeler.py

@@ -160,7 +160,7 @@ def accumulate_screenings(df: pd.DataFrame, settings: dict) -> pd.DataFrame:
                 number_session += 1
                 df.loc[row_week[0], 'HasCompletedSession'] = 1
                 df.loc[row_week[0], 'Baseline'] = 1
-                if row_mean_evaluation[1] > 4:
+                if row_mean_evaluation[1] >= 3.7:
                     df.loc[row_week[0], 'GotComplianceInSession'] = 1
                 else:
                     df.loc[row_week[0], 'GotComplianceInSession'] = 0