Skip to content

Commit

Permalink
Bugfix for binary class modeling.
Browse files Browse the repository at this point in the history
  • Loading branch information
@Reiter89
Reiter89 committed Apr 8, 2022
1 parent 9710e81 commit 9a07dee
Showing 1 changed file with 35 additions and 29 deletions.
64 changes: 35 additions & 29 deletions supplementcode/scan.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -5458,17 +5458,17 @@ def permutation_test(self, x, y, nperms=100, cv_method=model_selection.KFold(7,
perm_testmisclassifiedsamples = list()

if self.n_classes == 2:
perm_trainroc_fpr = numpy.zeros((nperms, len(y),))
perm_trainroc_fpr = numpy.zeros((nperms, len(y), 1))
perm_trainroc_fpr[:] = numpy.nan
perm_trainroc_tpr = numpy.zeros((nperms, len(y),))
perm_trainroc_tpr = numpy.zeros((nperms, len(y), 1))
perm_trainroc_tpr[:] = numpy.nan
perm_trainauc = numpy.zeros((nperms,))
perm_trainauc = numpy.zeros((nperms, 1))

perm_testroc_fpr = numpy.zeros((nperms, len(y),))
perm_testroc_fpr = numpy.zeros((nperms, len(y), 1))
perm_testroc_fpr[:] = numpy.nan
perm_testroc_tpr = numpy.zeros((nperms, len(y),))
perm_testroc_tpr = numpy.zeros((nperms, len(y), 1))
perm_testroc_tpr[:] = numpy.nan
perm_testauc = numpy.zeros((nperms,))
perm_testauc = numpy.zeros((nperms, 1))
else:
perm_trainroc_fpr = numpy.zeros((nperms, len(y), y_nvars))
perm_trainroc_fpr[:] = numpy.nan
Expand Down Expand Up @@ -5530,20 +5530,20 @@ def permutation_test(self, x, y, nperms=100, cv_method=model_selection.KFold(7,

if n_classes == 2:
# ROC AUC
fpr_train_mean = numpy.nanmean(permute_class.cvParameters['DA']['CV_TrainROC_fpr'].T[0], axis = 0)
tpr_train_mean = numpy.nanmean(permute_class.cvParameters['DA']['CV_TrainROC_tpr'].T[0], axis = 0)
fpr_train_mean = numpy.nanmean(permute_class.cvParameters['DA']['CV_TrainROC_fpr'], axis = 0)
tpr_train_mean = numpy.nanmean(permute_class.cvParameters['DA']['CV_TrainROC_tpr'], axis = 0)
auc_train_mean = numpy.nanmean(permute_class.cvParameters['DA']['CV_TrainAUC'], axis = 0)

fpr_test_mean = numpy.nanmean(permute_class.cvParameters['DA']['CV_TestROC_fpr'].T[0], axis = 0)
tpr_test_mean = numpy.nanmean(permute_class.cvParameters['DA']['CV_TestROC_tpr'].T[0], axis = 0)
fpr_test_mean = numpy.nanmean(permute_class.cvParameters['DA']['CV_TestROC_fpr'], axis = 0)
tpr_test_mean = numpy.nanmean(permute_class.cvParameters['DA']['CV_TestROC_tpr'], axis = 0)
auc_test_mean = numpy.nanmean(permute_class.cvParameters['DA']['CV_TestAUC'], axis = 0)

perm_trainroc_fpr[permutation, :len(fpr_train_mean)] = fpr_train_mean
perm_trainroc_tpr[permutation, :len(tpr_train_mean)] = tpr_train_mean
perm_trainroc_fpr[permutation, :len(fpr_train_mean), 0] = fpr_train_mean
perm_trainroc_tpr[permutation, :len(tpr_train_mean), 0] = tpr_train_mean
perm_trainauc[permutation] = auc_train_mean

perm_testroc_fpr[permutation, :len(fpr_test_mean)] = fpr_test_mean
perm_testroc_tpr[permutation, :len(tpr_test_mean)] = tpr_test_mean
perm_testroc_fpr[permutation, :len(fpr_test_mean), 0] = fpr_test_mean
perm_testroc_tpr[permutation, :len(tpr_test_mean), 0] = tpr_test_mean
perm_testauc[permutation] = auc_test_mean
else:
# ROC AUC
Expand Down Expand Up @@ -5753,17 +5753,17 @@ def bootstrap_test(self, x, y, nboots=100, stratify = True, outputdist=False, **
boot_testmisclassifiedsamples = list()

if self.n_classes == 2:
boot_trainroc_fpr = numpy.zeros((nboots, len(y),))
boot_trainroc_fpr = numpy.zeros((nboots, len(y), 1))
boot_trainroc_fpr[:] = numpy.nan
boot_trainroc_tpr = numpy.zeros((nboots, len(y),))
boot_trainroc_tpr = numpy.zeros((nboots, len(y), 1))
boot_trainroc_tpr[:] = numpy.nan
boot_trainauc = numpy.zeros((nboots,))
boot_trainauc = numpy.zeros((nboots, 1))

boot_testroc_fpr = numpy.zeros((nboots, len(y),))
boot_testroc_fpr = numpy.zeros((nboots, len(y), 1))
boot_testroc_fpr[:] = numpy.nan
boot_testroc_tpr = numpy.zeros((nboots, len(y),))
boot_testroc_tpr = numpy.zeros((nboots, len(y), 1))
boot_testroc_tpr[:] = numpy.nan
boot_testauc = numpy.zeros((nboots,))
boot_testauc = numpy.zeros((nboots, 1))
else:
boot_trainroc_fpr = numpy.zeros((nboots, len(y), y_nvars))
boot_trainroc_fpr[:] = numpy.nan
Expand Down Expand Up @@ -5875,8 +5875,6 @@ def bootstrap_test(self, x, y, nboots=100, stratify = True, outputdist=False, **
class_score = ChemometricsPLS.predict(bootstrap_pipeline, xtest)

# DA parameter


if n_classes == 2:
boot_testaccuracy[boot_i] = metrics.accuracy_score(ytest, y_pred)
boot_testprecision[boot_i] = metrics.precision_score(ytest, y_pred)
Expand All @@ -5886,17 +5884,17 @@ def bootstrap_test(self, x, y, nboots=100, stratify = True, outputdist=False, **
boot_testmatthews_mcc[boot_i] = metrics.matthews_corrcoef(ytest, y_pred)

# ROC AUC
fpr_mean = bootstrap_pipeline.modelParameters['DA']['ROC_fpr'].T[0]
tpr_mean = bootstrap_pipeline.modelParameters['DA']['ROC_tpr'].T[0]
fpr_mean = bootstrap_pipeline.modelParameters['DA']['ROC_fpr']
tpr_mean = bootstrap_pipeline.modelParameters['DA']['ROC_tpr']
auc_mean = bootstrap_pipeline.modelParameters['DA']['AUC']

boot_trainroc_fpr[boot_i, :len(fpr_mean)] = fpr_mean
boot_trainroc_tpr[boot_i, :len(tpr_mean)] = tpr_mean
boot_trainauc[boot_i] = auc_mean
boot_trainroc_fpr[boot_i, :len(fpr_mean), 0] = fpr_mean[:,0]
boot_trainroc_tpr[boot_i, :len(tpr_mean), 0] = tpr_mean[:,0]
boot_trainauc[boot_i] = auc_mean[0]

fpr, tpr, _ = metrics.roc_curve(ytest, class_score.ravel(), drop_intermediate = False)
boot_testroc_fpr[boot_i, :len(fpr)] = fpr
boot_testroc_tpr[boot_i, :len(tpr)] = tpr
boot_testroc_fpr[boot_i, :len(fpr), 0] = fpr
boot_testroc_tpr[boot_i, :len(tpr), 0] = tpr
boot_testauc[boot_i] = metrics.auc(fpr, tpr)

else:
Expand Down Expand Up @@ -9011,6 +9009,10 @@ def scan_mv_pls_rocauc(model, labels_group):
# Preallocate
dict_rocauc = {}

# Account for two classes
if len(labels_group) == 2:
labels_group = ["0"]

# Train
array_train_fpr_mean = numpy.nanmean(model.bootstrapParameters['DA']['boot_TrainROC_fpr'], axis = 0)
array_train_tpr_mean = numpy.nanmean(model.bootstrapParameters['DA']['boot_TrainROC_tpr'], axis = 0)
Expand Down Expand Up @@ -9844,6 +9846,10 @@ def scan_mv_pls_roc_plot_full(dict_roc, labels_group, inp):
df_test_roc = dict_roc['TestROC'].copy()
df_test_auc = dict_roc['TestAUC'].copy()

# Account for two classes
if len(labels_group) == 2:
labels_group = ["0"]

for item in labels_group:
auc_train_mean = numpy.round(df_train_auc.at[0, item+'_auc_mean'],5)
auc_test_mean = numpy.round(df_test_auc.at[0, item+'_auc_mean'],5)
Expand Down

0 comments on commit 9a07dee

Please sign in to comment.