Functions — github.com/davidstutz/confidence-calibrated-adversarial-training

↓ 560 callersMethoddistributionAt

(self, label, confidence, labels=10)

tests/test_eval.py:12

↓ 171 callersMethodconfidence_at_tpr

Confidence threshold for given true positive rate. :param tpr: true positive rate in [0, 1] :type tpr: float :return

common/eval/adversarial_evaluation.py:249

↓ 114 callersMethodconfidence_at_tpr

Confidence threshold for given true positive rate. :param tpr: true positive rate in [0, 1] ;type tpr: float :return

common/eval/distal_evaluation.py:179

↓ 87 callersMethodadd_scalar

Add scalar value. :param tag: tag for scalar :type tag: str :param value: value :type value: mixed :

common/summary.py:55

↓ 73 callersMethodconfidence_at_tpr

Confidence threshold for given true positive rate. :param tpr: true positive rate in [0, 1] ;type tpr: float :return

common/eval/corrupted_evaluation.py:188

↓ 70 callersMethoddetach

Detach a key. :param key: log file name :type key: str

common/log.py:179

↓ 64 callersMethodfpr_at_confidence

False positive rate at confidence threshold. :param threshold: confidence threshold in [0, 1] :type threshold: float

common/eval/adversarial_evaluation.py:304

↓ 64 callersMethodtpr_at_confidence

True positive rate at confidence threshold. :param threshold: confidence threshold in [0, 1] :type threshold: float

common/eval/adversarial_evaluation.py:275

↓ 62 callersMethodtest_error_at_confidence

Test error for given confidence threshold. :param threshold: confidence threshold :type threshold: float :return tes

common/eval/adversarial_evaluation.py:350

↓ 58 callersMethodrobust_test_error_at_confidence

Robust test error for given confidence threshold. :param threshold: confidence threshold :type threshold: float :ret

common/eval/adversarial_evaluation.py:428

↓ 51 callersFunctionlog

Quick access to logger instance. :param message: message or variable to log :type message: mixed :param level: level, i.e. color

common/log.py:244

↓ 50 callersMethodappend_layer

Add a layer. :param name: layer name :type name: str :param layer: layer :type layer: torch.nn.Module

models/classifier.py:90

↓ 45 callersMethodconfidence_at_tpr

Confidence threshold for given true positive rate. :param tpr: true positive rate in [0, 1] :type tpr: float :return

common/eval/clean_evaluation.py:98

↓ 44 callersMethodfpr_at_confidence

False positive rate at confidence threshold. :param threshold: confidence threshold in [0, 1] :type threshold: float

common/eval/distal_evaluation.py:282

↓ 44 callersMethodtpr_at_confidence

True positive rate at confidence threshold. :param threshold: confidence threshold in [0, 1] :type threshold: float

common/eval/distal_evaluation.py:205

↓ 42 callersMethodrunTestAttackPerformance

(self, attack, attempts=5, objective=attacks.objectives.UntargetedF0Objective())

tests/test_attacks_normal.py:104

↓ 40 callersMethodadd_histogram

Add histogram data. :param tag: tag :type tag: str :param values: values :type values: torch.Tensor or numpy

common/summary.py:87

↓ 35 callersMethodload

Load a state. :param filepath: file to load :type filepath: str :return: state :rtype: State

common/state.py:96

↓ 32 callersMethodtest_error_at_confidence

Test error for given confidence threshold. :param threshold: confidence threshold :type threshold: float :return tes

common/eval/corrupted_evaluation.py:265

↓ 31 callersMethodcorrupted_test_error_at_confidence

Test error for given confidence threshold. :param threshold: confidence threshold :type threshold: float :return tes

common/eval/corrupted_evaluation.py:313

↓ 31 callersMethodtest

Test step. :param epoch: epoch :type epoch: int

common/train/normal_training.py:145

↓ 30 callersMethodtpr_at_confidence

True positive rate at confidence threshold. :param threshold: confidence threshold in [0, 1] :type threshold: float

common/eval/clean_evaluation.py:121

↓ 29 callersMethodstep

Training + test step. :param epoch: epoch :type epoch: int

common/train/training_interface.py:65

↓ 27 callersMethodadd_text

Add images. :param tag: tag :type tag: str :param text_string: test :type text_string: str :param gl

common/summary.py:161

↓ 26 callersMethodtest_error

Test error. :return: test error :rtype: float

common/eval/adversarial_evaluation.py:340

↓ 25 callersFunctiondata_file

Generate path to data file. :param name: name of file :type name: str :param ext: extension (including period) :type ext: str

common/paths.py:38

↓ 24 callersMethodfpr_at_confidence

False positive rate at confidence threshold. :param threshold: confidence threshold in [0, 1] :type threshold: float

common/eval/corrupted_evaluation.py:351

↓ 24 callersMethodrobust_test_error

Robust test error. :return: robust test error :rtype: float

common/eval/adversarial_evaluation.py:388

↓ 24 callersMethodtest_error

Test error. :return: test error :rtype: float

common/eval/clean_evaluation.py:158

↓ 24 callersMethodtpr_at_confidence

True positive rate at confidence threshold. :param threshold: confidence threshold in [0, 1] :type threshold: float

common/eval/corrupted_evaluation.py:214

↓ 22 callersMethodmain

Main.

experiments/train.py:50

↓ 21 callersMethodrunTestAttackProjections

(self, attack, epsilon, objective=attacks.objectives.UntargetedF0Objective())

tests/test_attacks_normal.py:88

↓ 19 callersFunction__attack_config

Attack configuration. :param directory: directory for perturbations :type directory: str :param attack: attack :type attack: att

experiments/config/common.py:1092

↓ 18 callersMethodforward

(self, x_nat, y_nat)

attacks/batch_corner_search.py:172

↓ 17 callersFunction__untargeted_f0

Untargeted loss objective, default is cross-entropy. :param loss: loss to maximize :type loss: callable :return: objective :rtyp

experiments/config/common.py:883

↓ 15 callersMethodcheckpoint

Quick access to State.save. :param filepath: path to file :type filepath: str :param model: model :type mode

common/state.py:76

↓ 15 callersMethodflush

Flush.

common/summary.py:177

↓ 14 callersMethodset

Set true or target classes. Target classes might be None for untargeted attacks. :param true_classes: true classes :type tru

attacks/objectives.py:22

↓ 13 callersMethodcorrupted_test_error

Test error. :return: test error :rtype: float

common/eval/corrupted_evaluation.py:303

↓ 12 callersMethodrun

Run attack. :param model: model to attack :type model: torch.nn.Module :param images: images :type images: t

attacks/attack.py:19

↓ 12 callersMethodtest_error

Test error. :return: test error :rtype: float

common/eval/corrupted_evaluation.py:255

↓ 11 callersMethod__init__

(self, indices=None, resize=None)

common/datasets.py:431

↓ 9 callersMethodsave

Save the state. :param filepath: file to save to :type filepath: str

common/state.py:37

↓ 8 callersMethodsuccess

Get success. :param logits: logit output of the network :type logits: torch.autograd.Variable :return: success

attacks/objectives.py:57

↓ 8 callersMethodtest_error_at_confidence

Test error for given confidence threshold. :param threshold: confidence threshold :type threshold: float :return tes

common/eval/clean_evaluation.py:168

↓ 7 callersMethodget_instance

Get current log instance, simple singleton. :return: log :rtype: Log

common/log.py:204

↓ 7 callersMethodprogress

Report progress. :param epoch: epoch :type epoch: int :param batch: batch :type batch: int :param ba

common/train/training_interface.py:26

↓ 7 callersMethodreceiver_operating_characteristic_auc

Computes the ROC curve for correct classified vs. incorrect classified. :return: ROC AUC score :rtype: float

common/eval/clean_evaluation.py:216

↓ 7 callersMethodrunTestAttackPerformance

(self, attack, attempts=5)

tests/test_distal_attacks_normal.py:82

↓ 7 callersFunctionsave_latex

Save latex file. :param filepath: path to file :rtype filepath: str :param latex: latex code :type latex: str

experiments/eval/misc.py:522

↓ 7 callersMethodtarget_confidence

True confidence. :param logits: logit output of the network :type logits: torch.autograd.Variable :return: true conf

attacks/objectives.py:88

↓ 7 callersMethodtrue_confidence

True confidence. :param logits: logit output of the network :type logits: torch.autograd.Variable :return: true conf

attacks/objectives.py:72

↓ 7 callersMethodvalidate

Check validity.

common/experiments.py:410

↓ 6 callersMethodadd_images

Add images. :param tag: tag :type tag: str :param img_tensor: images :type img_tensor: torch.Tensor or numpy

common/summary.py:125

↓ 6 callersMethodrunTestAttackPerformance

(self, attack, attempts=5, objective=attacks.objectives.UntargetedF0Objective())

tests/test_attacks_adversarial.py:94

↓ 6 callersMethodtestTrainModel

(self, model)

tests/test_train_models.py:27

↓ 6 callersMethodupdate_queue

Update queue.

common/summary.py:587

↓ 5 callersMethod__init__

Constructor. :param args: shape :type args: [int]

common/torch.py:701

↓ 5 callersMethodalternative_robust_test_error_at_confidence

Robust test error for given confidence threshold. :param threshold: confidence threshold :type threshold: float :ret

common/eval/adversarial_evaluation.py:398

↓ 5 callersMethodfpr_at_99tpr

Test error at 95%TPR. :return: robust test error :rtype: float

common/eval/clean_evaluation.py:382

↓ 5 callersMethodfpr_at_confidence

False positive rate at confidence threshold. :param threshold: confidence threshold in [0, 1] :type threshold: float

common/eval/clean_evaluation.py:146

↓ 5 callersMethodreceiver_operating_characteristic_auc

Computes the ROC curve for correct classified vs. incorrect classified. :return: ROC AUC score :rtype: float

common/eval/adversarial_evaluation.py:550

↓ 5 callersMethodrobust_test_error_at_99tpr

Robust test error at 95%TPR. :return: robust test error :rtype: float

common/eval/adversarial_evaluation.py:876

↓ 4 callersMethod__init__

Constructor.

common/experiments.py:393

↓ 4 callersMethod_make_layer

(self, block, planes, num_blocks, stride)

models/preact_resnet.py:65

↓ 4 callersMethodconfidence_at_99tpr

Test error at 95%TPR. :return: robust test error :rtype: float

common/eval/clean_evaluation.py:262

↓ 4 callersFunctionexpand_as

Expands the tensor using view to allow broadcasting. :param tensor: input tensor :type tensor: torch.Tensor or torch.autograd.Variable

common/torch.py:437

↓ 4 callersMethodfmargin

(self, x, y)

attacks/batch_cube2.py:12

↓ 4 callersMethodlog

Log a message. :param message: message or variable to log :type message: mixed :param level: level, i.e. color

common/log.py:226

↓ 4 callersFunctionp_selection

The schedule was adapted for mnist/cifar10, but not for imagenet

attacks/batch_cube2.py:22

↓ 4 callersMethodrunTestDeterministicAttack

(self, attack, objective=attacks.objectives.UntargetedF0Objective())

tests/test_attacks_normal.py:122

↓ 4 callersMethodsuccess_rate

Success rate. :return: success rate :rtype: float

common/eval/adversarial_evaluation.py:499

↓ 4 callersMethodsuccess_rate_at_confidence

Success rate at confidence threshold. :param threshold: confidence threshold :type threshold: float :return: success

common/eval/adversarial_evaluation.py:509

↓ 4 callersMethodtest_error_at_confidence

Test error for given confidence threshold. :param threshold: confidence threshold :type threshold: float :return tes

common/eval/distal_evaluation.py:244

↓ 4 callersMethodtimestamp

Print timestamp. :return: date and time :rtype: str

common/log.py:88

↓ 4 callersMethodtnr_at_confidence

True negative rate at confidence threshold. :param threshold: confidence threshold in [0, 1] :type threshold: float

common/eval/corrupted_evaluation.py:226

↓ 4 callersMethodtnr_at_confidence

False positive rate at confidence threshold. :param threshold: confidence threshold in [0, 1] :type threshold: float

common/eval/adversarial_evaluation.py:328

↓ 4 callersMethodtnr_at_confidence

False positive rate at confidence threshold. :param threshold: confidence threshold in [0, 1] :type threshold: float

common/eval/distal_evaluation.py:294

↓ 4 callersMethodvalidation_tpr_at_confidence

True positive rate at confidence threshold. :param threshold: confidence threshold in [0, 1] :type threshold: float

common/eval/corrupted_evaluation.py:238

↓ 4 callersMethodvalidation_tpr_at_confidence

True positive rate at confidence threshold. :param threshold: confidence threshold in [0, 1] :type threshold: float

common/eval/adversarial_evaluation.py:287

↓ 4 callersMethodvalidation_tpr_at_confidence

True positive rate at confidence threshold. :param threshold: confidence threshold in [0, 1] :type threshold: float

common/eval/clean_evaluation.py:133

↓ 4 callersMethodvalidation_tpr_at_confidence

True positive rate at confidence threshold. :param threshold: confidence threshold in [0, 1] :type threshold: float

common/eval/distal_evaluation.py:217

↓ 3 callersMethod__init__

(self, min=0, max=1, name=None, deterministic=False, random_state=None)

common/imgaug.py:49

↓ 3 callersMethod_files

Get files. :return: files :rtype: [File]

common/log.py:193

↓ 3 callersMethod_wide_layer

(self, block, planes, blocks, stride)

models/wide_resnet.py:104

↓ 3 callersMethodelapsed

Get elapsed time in seconds :return: elapsed time in seconds :rtype: float

common/timer.py:24

↓ 3 callersFunctionflat2square

returns the position and the perturbation given the index of an image of the batch of all the possible perturbations

attacks/batch_corner_search.py:72

↓ 3 callersMethodgetModelFile

(cls)

tests/test_attacks_normal.py:69

↓ 3 callersMethodgetModelFile

(cls)

tests/test_distal_attacks_normal.py:65

↓ 3 callersMethodgetModelFile

(cls)

tests/test_attacks_adversarial.py:75

↓ 3 callersFunctionis_cuda

Check if model/tensor is on CUDA. :param mixed: model or tensor :type mixed: torch.nn.Module or torch.autograd.Variable or torch.Tensor

common/torch.py:46

↓ 3 callersMethodkwargs_get

Get argument if not None. :param key: key :type key: str :param default: default value :type default: mixed

models/classifier.py:73

↓ 3 callersFunctionmakedir

Creates directory if it does not exist. :param dir: directory path :type dir: str

common/utils.py:210

↓ 3 callersMethodnormalize

Normalization. :param gradients: gradients :type gradients: torch.autograd.Variable

attacks/norms.py:20

↓ 3 callersMethodreceiver_operating_characteristic_auc

Computes the ROC curve for correct classified vs. incorrect classified. :return: ROC AUC score :rtype: float

common/eval/distal_evaluation.py:325

↓ 3 callersMethodrun

Run attack. :param model: model to attack :type model: torch.nn.Module :param images: images :type images: t

attacks/batch_gradient_descent.py:54

↓ 3 callersMethodtest_error

Test error. :return: test error :rtype: float

common/eval/distal_evaluation.py:234

↓ 3 callersMethodtest_error_at_99tpr

Test error at 95%TPR. :return: robust test error :rtype: float

common/eval/adversarial_evaluation.py:796

↓ 2 callersFunction__adversarial_training

Adversarial training configuration. :param directory: directory for models :type directory: str :param attack: attack for adversaria

experiments/config/common.py:986

↓ 2 callersMethod__init__

Constructor.

attacks/objectives.py:11

Functions916 in github.com/davidstutz/confidence-calibrated-adversarial-training