fit

load_dataset.py

@@ -10,7 +10,7 @@ def safe_float(x):
 
 
 
-def analisis_univariado(dfi, target=None, continuas=[], discretas=[]):
+def univariate_analysis(dfi, target=None, continous=[], discrete=[]):
     if target is None:
         raise ValueError("No target variable provided")
 
@@ -33,8 +33,8 @@ def analisis_univariado(dfi, target=None, continuas=[], discretas=[]):
     results = []
     resultsmody = []
 
-    # Análisis de variables continuas
-    for var in continuas:
+    # Análisis de variables continous
+    for var in continous:
         Xvar = dfi[var].T
         group1_values = data_group1[var].T
         group2_values = data_group2[var].T
@@ -82,8 +82,8 @@ def analisis_univariado(dfi, target=None, continuas=[], discretas=[]):
                 f"{mw_pval:.3f}", ("*" if mw_pval < 0.05 else "NS"), f"{mediang:.1f} ({qlg:.1f} - {qrg:.1f})"
             ])
 
-    # Análisis de variables discretas
-    for var in discretas:
+    # Análisis de variables discrete
+    for var in discrete:
         freq_table = dfi.groupby([target, var]).size().unstack(fill_value=0)
         percentages = freq_table.div(freq_table.sum(axis=1), axis=0) * 100
 

train.py

@@ -1,28 +1,53 @@
-from load_dataset import load_data, analisis_univariado
+from load_dataset import load_data, univariate_analysis
 #from trainer import BinaryTuner
 import pandas as pd
 import numpy as np
 import warnings
 warnings.filterwarnings("ignore")
 
+# Load data from HC.xslc raw original data file and extract mody data into filtered dataframes
+# filtered data is stord as `MODY_data.xlsx`  for screening.
+# this returns a tuple  (mody1, mody2, mody3, mody5) dataFrames from each sheet in the filtered record
 _, dms2, dms3, _ = load_data()
 
-
-with pd.ExcelWriter("Univariado2.xlsx", engine='xlsxwriter') as xls:
-    mody2 = analisis_univariado(dms2.dropna(), target="MODY2_label", continuas=['edad diag', 'IMC', 'glu ayu', 'glu 120','A1c'], discretas=['sexo', 'diabetes_familia'])
+# Univariate analysis of the origin data
+with pd.ExcelWriter("UnivariateODY.xlsx", engine='xlsxwriter') as xls:
+    # continous are tested shapiro for normality. metrics for groups are computed, then ttested or mannwhitneyu for difference in groups
+    # binary use fisher and frecuency tables
+    mody2 = univariate_analysis(dms2.dropna(), target="MODY2_label", continous=['edad diag', 'IMC', 'glu ayu', 'glu 120','A1c'], discrete=['sexo', 'diabetes_familia'])
     mody2.to_excel(xls, sheet_name='MODY2', index=False)
 
-    mody3 = analisis_univariado(dms3.dropna(), target="MODY3_label", continuas=['edad diag', 'IMC', 'glu ayu', 'glu 120','A1c'], discretas=['sexo', 'diabetes_familia'])
+    mody3 = univariate_analysis(dms3.dropna(), target="MODY3_label", continous=['edad diag', 'IMC', 'glu ayu', 'glu 120','A1c'], discrete=['sexo', 'diabetes_familia'])
     mody3.to_excel(xls, sheet_name='MODY3', index=False)
 
+# For reproducibility, seeds from the original research, set to None to get random
+seeds = [231964, 48928, 132268, 113986, 574626, 130068, 226585, 446306, 535997, 685636, 779992, 600946, 231614, 1027776, 747054, 546372, 885843, 536202, 852539, 848580, 997648, 440679, 118304, 49131, 861767]
+# seeds = None
+# or
+# import np
+# seeds = [np.random.randint(1, 2**20) for _ in range(n_seeds)]
 
-# mody2 = BinaryTuner(dms2, 'MODY2_label', seeds=[231964], test_size=0.2)
-# mody2.fit()
-# mody2.explain_model('GaussianNB', 'fulldataset-oversampled-mice', 231964)
-# mody2.wrap_and_save()
-#
-#
-# mody3 = BinaryTuner(dms3, 'MODY3_label', seeds=[536202], test_size=0.2)
-# mody3.fit()
-# mody3.explain_model('RandomForestClassifier', 'fulldataset-original-mice', 536202)
-# mody3.wrap_and_save()
+# The BinaryTuner do a Grid Search for multiple hyperparameter space and
+# save metrics the test split for the best combinations of ML hyperparameters
+#  * Hyperparameters GridSearch for each ML Model for up to 60 different combinations
+#  * 10 different Machine Learning Models capable of Binary Clasification
+#  * Model trained on data with no missing values, and impute MICE and KNN
+#  * Different random train and test splits, for given test_size ratio
+# A folder is created with the label name with all the state and run data
+mody2 = BinaryTuner(dms2, 'MODY2_label', seeds=seed, test_size=0.2)
+
+# CheckPoint during search all trained models are saved inside the label folder
+# including the scalers required for inference and testing
+mody2.fit()
+
+# Generate Chap Explainer for (Model, dataset, seed). See folder structure.
+mody2.explain_model('GaussianNB', 'fulldataset-oversampled-mice', 231964)
+
+# Create summary of metrics and zip the folder for easy portabilitys
+mody2.wrap_and_save()
+
+# Repeat process for next dataset
+mody3 = BinaryTuner(dms3, 'MODY3_label', seeds=seeds, test_size=0.2)
+mody3.fit()
+mody3.explain_model('RandomForestClassifier', 'fulldataset-original-mice', 536202)
+mody3.wrap_and_save()

trainer.py

@@ -599,8 +599,8 @@ class BinaryTuner:
                 self.saveCheckPoint()
         self.bar.close()
 
-    def get_best_models(self):
-        return self.ledger.groupby(["Dataset", "Model"])["ROC_AUC"].agg(['mean', 'std'])
+    def get_best_models(self, metric="ROC_AUC"):
+        return self.ledger.groupby(["Dataset", "Model"])[metric].agg(['mean', 'std'])
 
     def explain_model(self, modelname=None, dataset=None, seed=None):
         self.logger.info("{:=^60}".format(' Begin SHAP Explainer: {} {} {} '.format(modelname, dataset, seed)))
@@ -743,9 +743,9 @@ class BinaryTuner:
 
     def wrap_and_save(self):
         self.logger.info("{:=^60}".format(' Saving Summary and Wrap the output in a ZipFile '))
-
-        with pd.ExcelWriter('{}/Summary.xlsx'.format(self.name) , engine='xlsxwriter') as xls:
-            self.get_best_models().to_excel(xls, sheet_name='Results')
+        for metric in ["ROC_AUC", "NPV", "PPV", "Brier", "sensitivity", "specificity"]:
+            with pd.ExcelWriter('{}/Summary-{}.xlsx'.format(self.name, metric) , engine='xlsxwriter') as xls:
+                self.get_best_models(metric).to_excel(xls, sheet_name='Results')
 
         with zipfile.ZipFile('{}-{}.zip'.format(self.name, self.start), 'w', zipfile.ZIP_DEFLATED) as zipf:
             for root, dirs, files in os.walk(self.name):