Upload app.py

app.py

@@ -140,88 +140,47 @@ unique_PRIMARYMETHODPAYMENT = ['Private/commercial insurance', 'Medicaid', 'Medi
 #Prepare data for the outcome 1 (mortality).
 y1 = x1.pop('OUTCOME')
 categorical_columns1 = list(x1.select_dtypes('object').columns)
-x1 = x1.astype({col: "category" for col in categorical_columns1})
-y1_data_xgb = xgb.DMatrix(x1, label=y1, enable_categorical=True)
-x1_lgb = x1.rename(columns = lambda x:re.sub('[^A-Za-z0-9_]+', '', x))
-y1_data_lgb = lgb.Dataset(x1_lgb, label=y1) 
-y1_data_cb = Pool(data=x1, label=y1, cat_features=categorical_columns1)
-x1_rf = x1
-categorical_columns1 = list(x1_rf.select_dtypes('category').columns)
-x1_rf = x1_rf.astype({col: "category" for col in categorical_columns1})
 le = sklearn.preprocessing.LabelEncoder()
 for col in categorical_columns1:
-        x1_rf[col] = le.fit_transform(x1_rf[col].astype(str))
-d1 = dict.fromkeys(x1_rf.select_dtypes(np.int64).columns, str)
-x1_rf = x1_rf.astype(d1)
+        x1[col] = le.fit_transform(rf[col].astype(str))
+d1 = dict.fromkeys(x1.select_dtypes(np.int64).columns, str)x1_rf = x1_rf.astype(d1)
+x1 = x1.astype(d1)
 
 #Prepare data for the outcome 2 (discharge).
 y2 = x2.pop('OUTCOME')
 categorical_columns2 = list(x2.select_dtypes('object').columns)
-x2 = x2.astype({col: "category" for col in categorical_columns2})
-y2_data_xgb = xgb.DMatrix(x2, label=y2, enable_categorical=True)
-x2_lgb = x2.rename(columns = lambda x:re.sub('[^A-Za-z0-9_]+', '', x))
-y2_data_lgb = lgb.Dataset(x2_lgb, label=y2) 
-y2_data_cb = Pool(data=x2, label=y2, cat_features=categorical_columns2)
-x2_rf = x2
-categorical_columns2 = list(x2_rf.select_dtypes('category').columns)
-x2_rf = x2_rf.astype({col: "category" for col in categorical_columns2})
 le = sklearn.preprocessing.LabelEncoder()
 for col in categorical_columns2:
-        x2_rf[col] = le.fit_transform(x2_rf[col].astype(str))
-d2 = dict.fromkeys(x2_rf.select_dtypes(np.int64).columns, str)
-x2_rf = x2_rf.astype(d2)
+        x2[col] = le.fit_transform(rf[col].astype(str))
+d2 = dict.fromkeys(x2.select_dtypes(np.int64).columns, str)x2_rf = x2_rf.astype(d2)
+x2 = x2.astype(d2)
 
 #Prepare data for the outcome 3 (LOS).
 y3 = x3.pop('OUTCOME')
 categorical_columns3 = list(x3.select_dtypes('object').columns)
-x3 = x3.astype({col: "category" for col in categorical_columns3})
-y3_data_xgb = xgb.DMatrix(x3, label=y3, enable_categorical=True)
-x3_lgb = x3.rename(columns = lambda x:re.sub('[^A-Za-z0-9_]+', '', x))
-y3_data_lgb = lgb.Dataset(x3_lgb, label=y3) 
-y3_data_cb = Pool(data=x3, label=y3, cat_features=categorical_columns3)
-x3_rf = x3
-categorical_columns3 = list(x3_rf.select_dtypes('category').columns)
-x3_rf = x3_rf.astype({col: "category" for col in categorical_columns3})
 le = sklearn.preprocessing.LabelEncoder()
 for col in categorical_columns3:
-        x3_rf[col] = le.fit_transform(x3_rf[col].astype(str))
-d3 = dict.fromkeys(x3_rf.select_dtypes(np.int64).columns, str)
-x3_rf = x3_rf.astype(d3)
+        x3[col] = le.fit_transform(rf[col].astype(str))
+d3 = dict.fromkeys(x3.select_dtypes(np.int64).columns, str)x3_rf = x3_rf.astype(d3)
+x3 = x3.astype(d3)
 
 #Prepare data for the outcome 4 (ICU LOS).
 y4 = x4.pop('OUTCOME')
 categorical_columns4 = list(x4.select_dtypes('object').columns)
-x4 = x4.astype({col: "category" for col in categorical_columns4})
-y4_data_xgb = xgb.DMatrix(x4, label=y4, enable_categorical=True)
-x4_lgb = x4.rename(columns = lambda x:re.sub('[^A-Za-z0-9_]+', '', x))
-y4_data_lgb = lgb.Dataset(x4_lgb, label=y4) 
-y4_data_cb = Pool(data=x4, label=y4, cat_features=categorical_columns4)
-x4_rf = x4
-categorical_columns4 = list(x4_rf.select_dtypes('category').columns)
-x4_rf = x4_rf.astype({col: "category" for col in categorical_columns4})
 le = sklearn.preprocessing.LabelEncoder()
 for col in categorical_columns4:
-        x4_rf[col] = le.fit_transform(x4_rf[col].astype(str))
-d4 = dict.fromkeys(x4_rf.select_dtypes(np.int64).columns, str)
-x4_rf = x4_rf.astype(d4)
+        x4[col] = le.fit_transform(rf[col].astype(str))
+d4 = dict.fromkeys(x4.select_dtypes(np.int64).columns, str)x4_rf = x4_rf.astype(d4)
+x4 = x4.astype(d4)
 
 #Prepare data for the outcome 5 (complications).
 y5 = x5.pop('OUTCOME')
 categorical_columns5 = list(x5.select_dtypes('object').columns)
-x5 = x5.astype({col: "category" for col in categorical_columns5})
-y5_data_xgb = xgb.DMatrix(x5, label=y5, enable_categorical=True)
-x5_lgb = x5.rename(columns = lambda x:re.sub('[^A-Za-z0-9_]+', '', x))
-y5_data_lgb = lgb.Dataset(x5_lgb, label=y5) 
-y5_data_cb = Pool(data=x5, label=y5, cat_features=categorical_columns5)
-x5_rf = x5
-categorical_columns5 = list(x5_rf.select_dtypes('category').columns)
-x5_rf = x5_rf.astype({col: "category" for col in categorical_columns5})
 le = sklearn.preprocessing.LabelEncoder()
 for col in categorical_columns5:
-        x5_rf[col] = le.fit_transform(x5_rf[col].astype(str))
-d5 = dict.fromkeys(x5_rf.select_dtypes(np.int64).columns, str)
-x5_rf = x5_rf.astype(d5)
-
+        x5[col] = le.fit_transform(rf[col].astype(str))
+d5 = dict.fromkeys(x5.select_dtypes(np.int64).columns, str)x5_rf = x5_rf.astype(d5)
+x5 = x5.astype(d5)
 
 #Assign hyperparameters.
 y1_params = {'objective': 'binary:logistic', 'booster': 'gbtree', 'lambda': 0.5059844209148782, 'alpha': 0.0030156848979492556, 'max_depth': 2, 'eta': 4.546875002603483e-07, 'gamma': 1.1982641538268563e-08, 'grow_policy': 'lossguide', 'eval_metric': 'auc', 'verbosity': 0, 'seed': 31}
@@ -233,36 +192,47 @@ y5_params =  {'objective': 'binary', 'boosting_type': 'gbdt', 'lambda_l1': 0.001
 
 
 #Training models.
-y1_model_xgb = xgb.train(params = y1_params, dtrain = y1_data_xgb)
-y1_explainer_xgb = shap.TreeExplainer(y1_model_xgb)
+from xgboost import XGBClassifier
+y1_xgb = XGBClassifier(**y1_params)
+y1_model_xgb = y1_xgb.fit(x1, y1)
+y1_explainer_xgb = shap.TreeExplainer(y2_model_xgb)
 
 from sklearn.ensemble import RandomForestClassifier as rf
 y2_rf = rf(**y2_params)
-y2_model_rf = y2_rf.fit(x2_rf, y2)
+y2_model_rf = y2_rf.fit(x2, y2)
 y2_explainer_rf = shap.TreeExplainer(y2_model_rf)
 
-y3_model_xgb = xgb.train(params = y3_params, dtrain = y3_data_xgb)
-y3_explainer_xgb = shap.TreeExplainer(y1_model_xgb)
+from xgboost import XGBClassifier
+y3_xgb = XGBClassifier(**y3_params)
+y3_model_xgb = y3_xgb.fit(x3, y3)
+y3_explainer_xgb = shap.TreeExplainer(y3_model_xgb)
 
-y4_model_lgb = lgb.train(params = y4_params, train_set = y4_data_lgb)
+from lightgbm import LGBMClassifier
+lgb = LGBMClassifier(**y4_params)
+y4_model_lgb = lgb.fit(x4, y4)
 y4_explainer_lgb = shap.TreeExplainer(y4_model_lgb)
 
-y5_model_lgb = lgb.train(params=y5_params, train_set = y5_data_lgb)
+from lightgbm import LGBMClassifier
+lgb = LGBMClassifier(**y5_params)
+y5_model_lgb = lgb.fit(x5, y5)
 y5_explainer_lgb = shap.TreeExplainer(y5_model_lgb)
 
-
 #Define predict for y1 (mortality).
 def y1_predict_xgb(*args):
     df1 = pd.DataFrame([args], columns=x1.columns)
-    df1 = df1.astype({col: "category" for col in categorical_columns1})
-    pos_pred = y1_model_xgb.predict_proba(xgb.DMatrix(df1, enable_categorical=True))
-    return {"Mortality": float(pos_pred[0]), "No Mortality": 1 - float(pos_pred[0])}
+    df1 = df.astype({col: "category" for col in categorical_columns1})
+    d1 = dict.fromkeys(df1.select_dtypes(np.int64).columns, np.int32)
+    df1 = df1.astype(d1)
+    pos_pred = y1_model_xgb.predict_proba(df1)
+    return {"Mortality": float(pos_pred[0][1]), "No Mortality": float(pos_pred[0][0])}
 
 def y1_predict_lgb(*args):
-    df1 = pd.DataFrame([args], columns=x1_lgb.columns)
-    df1 = df1.astype({col: "category" for col in categorical_columns1})
-    pos_pred = y1_model_lgb.predict(df1)
-    return {"Mortality": float(pos_pred[0]), "No Mortality": 1 - float(pos_pred[0])}
+    df1 = pd.DataFrame([args], columns=x1.columns)
+    df1 = df.astype({col: "category" for col in categorical_columns1})
+    d1 = dict.fromkeys(df1.select_dtypes(np.int64).columns, np.int32)
+    df1 = df1.astype(d1)
+    pos_pred = y1_model_lgb.predict_proba(df1)
+    return {"Mortality": float(pos_pred[0][1]), "No Mortality": float(pos_pred[0][0])}
 
 def y1_predict_cb(*args):
     df1 = pd.DataFrame([args], columns=x1.columns)
@@ -274,7 +244,7 @@ def y1_predict_rf(*args):
     df1 = pd.DataFrame([args], columns=x1_rf.columns)
     df1 = df.astype({col: "category" for col in categorical_columns1})
     d1 = dict.fromkeys(df1.select_dtypes(np.int64).columns, np.int32)
-    d1f = df1.astype(d1)
+    df1 = df1.astype(d1)
     pos_pred = y1_model_rf.predict_proba(df1)
     return {"Mortality": float(pos_pred[0][1]), "No Mortality": float(pos_pred[0][0])}
 
@@ -282,7 +252,7 @@ def y1_predict_rf(*args):
 def y2_predict_xgb(*args):
     df2 = pd.DataFrame([args], columns=x2.columns)
     df2 = df2.astype({col: "category" for col in categorical_columns2})
-    pos_pred = y2_model_xgb.predict_proba(xgb.DMatrix(df2, enable_categorical=True))
+    pos_pred = y2_model_xgb.predict(xgb.DMatrix(df2, enable_categorical=True))
     return {"Facility Discharge": float(pos_pred[0]), "Home Discharge": 1 - float(pos_pred[0])}
 
 def y2_predict_lgb(*args):
@@ -309,7 +279,7 @@ def y2_predict_rf(*args):
 def y3_predict_xgb(*args):
     df3 = pd.DataFrame([args], columns=x3.columns)
     df3 = df3.astype({col: "category" for col in categorical_columns3})
-    pos_pred = y3_model_xgb.predict_proba(xgb.DMatrix(df3, enable_categorical=True))
+    pos_pred = y3_model_xgb.predict(xgb.DMatrix(df3, enable_categorical=True))
     return {"Prolonged LOS": float(pos_pred[0]), "No Prolonged LOS": 1 - float(pos_pred[0])}
 
 def y3_predict_lgb(*args):
@@ -336,7 +306,7 @@ def y3_predict_rf(*args):
 def y4_predict_xgb(*args):
     df4 = pd.DataFrame([args], columns=x4.columns)
     df4 = df4.astype({col: "category" for col in categorical_columns4})
-    pos_pred = y4_model_xgb.predict_proba(xgb.DMatrix(df4, enable_categorical=True))
+    pos_pred = y4_model_xgb.predict(xgb.DMatrix(df4, enable_categorical=True))
     return {"Prolonged ICU LOS": float(pos_pred[0]), "No Prolonged ICU LOS": 1 - float(pos_pred[0])}
 
 def y4_predict_lgb(*args):
@@ -363,7 +333,7 @@ def y4_predict_rf(*args):
 def y5_predict_xgb(*args):
     df5 = pd.DataFrame([args], columns=x5.columns)
     df5 = df5.astype({col: "category" for col in categorical_columns5})
-    pos_pred = y5_model_xgb.predict_proba(xgb.DMatrix(df5, enable_categorical=True))
+    pos_pred = y5_model_xgb.predict(xgb.DMatrix(df5, enable_categorical=True))
     return {"Major Complications": float(pos_pred[0]), "No Major Complications": 1 - float(pos_pred[0])}
 
 def y5_predict_lgb(*args):