Fix the random seed when bootstrap in Survival analysis evalution.

salan668 · Feb 5, 2024 · 12b94c6 · 12b94c6
1 parent 4321dd8
commit 12b94c6
Show file tree

Hide file tree

Showing 4 changed files with 24 additions and 14 deletions.
diff --git a/SA/FeatureSelector.py b/SA/FeatureSelector.py
@@ -118,12 +118,12 @@ def ClusterSelect(dc, number):
     sub_features = []
 
     if number == 1:
-        mylog.warning('The minimum number of KMeans is 2, Kmeans doesn\'t apply')
+        # mylog.warning('The minimum number of KMeans is 2, Kmeans doesn\'t apply')
         pccs = [abs(pearsonr(dc.df[one_feature].values, dc.event)[0]) for one_feature in dc.feature_name]
         selected_feature = dc.feature_name[pccs.index(max(pccs))]
         sub_features.append(selected_feature)
     else:
-        clusters = KMeans(n_clusters=number, random_state=0, init='k-means++').fit_predict(dc.array.transpose())
+        clusters = KMeans(n_clusters=number, random_state=0, init='k-means++', n_init="auto").fit_predict(dc.array.transpose())
 
         for i in range(number):
             clustering_features = [name for cluster_index, name in zip(clusters, dc.feature_name) if cluster_index == i]

diff --git a/SA/Fitter.py b/SA/Fitter.py
@@ -6,7 +6,6 @@
 import os
 import pickle
 import random
-from abc import abstractmethod
 from lifelines import CoxPHFitter, AalenAdditiveFitter
 
 from lifelines.utils.printer import Printer
@@ -15,6 +14,8 @@
 from SA.Utility import mylog
 from SA.DataContainer import DataContainer
 
+random.seed(0)
+
 
 class BaseFitter(object):
     def __init__(self, fitter=None, name=None):
@@ -48,9 +49,8 @@ def Summary(self):
 
 
 class CoxPH(BaseFitter):
-    def __init__(self):
-        random.seed(0)
-        super(CoxPH, self).__init__(CoxPHFitter(), self.__class__.__name__)
+    def __init__(self, penalizer=0.1):
+        super(CoxPH, self).__init__(CoxPHFitter(penalizer=penalizer), self.__class__.__name__)
 
     def Fit(self, dc: DataContainer):
         self.fitter.fit(dc.df, duration_col=dc.duration_name, event_col=dc.event_name)

diff --git a/SA/PipelineManager.py b/SA/PipelineManager.py
@@ -5,6 +5,7 @@
 """
 import os
 import csv
+
 import numpy as np
 import pandas as pd
 
@@ -183,10 +184,15 @@ def RunWithoutCV(self, train_dc, test_dc=None, store_folder=None):
                         fs_store_folder = MakeFolder(reduce_store_folder, '{}_{}'.format(
                             feature_selector.name, feature_number))
 
-                        feature_selector.selected_number = feature_number
+                        if feature_number >= len(dr_train_dc.GetFeatureName()):
+                            feature_selector.selected_number = len(dr_train_dc.GetFeatureName())
+                        else:
+                            feature_selector.selected_number = feature_number
+
                         feature_selector.Fit(dr_train_dc)
                         fs_train_dc = feature_selector.Transform(dr_train_dc, store_folder=fs_store_folder,
                                                                  store_key=TRAIN)
+
                         if test_dc is not None and not test_dc.IsEmpty():
                             fs_test_dc = feature_selector.Transform(dr_test_dc, store_folder=fs_store_folder,
                                                                     store_key=TEST)
@@ -259,10 +265,14 @@ def RunCV(self, dc, store_folder=None):
                             fs_store_folder = MakeFolder(reduce_store_folder, '{}_{}'.format(
                                 feature_selector.name, feature_number))
 
-                            feature_selector.selected_number = feature_number
-                            feature_selector.Fit(dr_cv_train_dc)
-                            fs_cv_train_dc = feature_selector.Transform(dr_cv_train_dc)
-                            fs_cv_val_dc = feature_selector.Transform(dr_cv_val_dc)
+                            if feature_number >= len(dr_cv_train_dc.GetFeatureName()):
+                                fs_cv_train_dc = dr_cv_train_dc
+                                fs_cv_val_dc = dr_cv_val_dc
+                            else:
+                                feature_selector.selected_number = feature_number
+                                feature_selector.Fit(dr_cv_train_dc)
+                                fs_cv_train_dc = feature_selector.Transform(dr_cv_train_dc)
+                                fs_cv_val_dc = feature_selector.Transform(dr_cv_val_dc)
 
                             for fitter_index, fitter in enumerate(self.fitters):
                                 fitter_store_folder = MakeFolder(fs_store_folder, fitter.name)

diff --git a/SA/Utility/Matric.py b/SA/Utility/Matric.py
@@ -4,7 +4,7 @@
 --2021/1/18
 """
 import numpy as np
-from random import choices
+import random
 from pycox.evaluation import EvalSurv
 
 from SA.Utility.Constant import *
@@ -17,11 +17,11 @@ def __init__(self, bootstrap_n=100):
         self.bootstrap_n = bootstrap_n
 
     def Bootstrap(self, surv, event: list, duration: list):
-        np.random.seed(42)  # control reproducibility
+        random.seed(42)  # control reproducibility
 
         cindex, brier, nbll = [], [], []
         for _ in range(self.bootstrap_n):
-            sampled_index = choices(range(surv.shape[1]), k=surv.shape[1])
+            sampled_index = random.choices(range(surv.shape[1]), k=surv.shape[1])
 
             sampled_surv = surv.iloc[:, sampled_index]
             sampled_event = [event[i] for i in sampled_index]