def test_precommit__BaseReverseNetwork_reverse_check_minmax():
def method(model):
return Gradient(model,
reverse_verbose=True,
reverse_check_min_max_values=True)
dryrun.test_analyzer(method, "mnist.*")
def test_fast__BaseReverseNetwork_reverse_check_minmax():
def method(model):
return Gradient(model,
reverse_verbose=True,
reverse_check_min_max_values=True)
dryrun.test_analyzer(method, "trivia.*:mnist.log_reg")
def test_imagenet__PatternAttribution():
def method(model):
# enough for test purposes, only pattern application is tested here
# pattern computation is tested separately.
# assume that one dim weights are biases, drop them.
patterns = [x for x in model.get_weights() if len(x.shape) > 1]
return PatternAttribution(model, patterns=patterns)
dryrun.test_analyzer(method, "imagenet.vgg16:imagenet.vgg19")
def test_fast__PatternNet():
def method(model):
# enough for test purposes, only pattern application is tested here
# pattern computation is tested separately.
# assume that one dim weights are biases, drop them.
patterns = [x for x in model.get_weights() if len(x.shape) > 1]
return PatternNet(model, patterns=patterns)
dryrun.test_analyzer(method, "mnist.log_reg")
def test_fast__AnalyzerNetworkBase_neuron_selection_index():
class CustomAnalyzer(Gradient):
def analyze(self, X):
index = 0
return super(CustomAnalyzer, self).analyze(X, index)
def method(model):
return CustomAnalyzer(model, neuron_selection_mode="index")
dryrun.test_analyzer(method, "trivia.*:mnist.log_reg")
def test_precommit__AnalyzerNetworkBase_neuron_selection_index():
class CustomAnalyzer(Gradient):
def analyze(self, X):
index = 3
return super(CustomAnalyzer, self).analyze(X, index)
def method(model):
return CustomAnalyzer(model, neuron_selection_mode="index")
dryrun.test_analyzer(method, "mnist.*")
def test_fast__DryRunAnalyzerTestCase():
"""
Sanity test for the TestCase.
"""
def method(output_layer):
class TestAnalyzer(object):
def analyze(self, X):
return X
return TestAnalyzer()
dryrun.test_analyzer(method, "trivia.*:mnist.log_reg")
def test_precommit__Gradient():
def method(model):
return Gradient(model)
dryrun.test_analyzer(method, "mnist.*")
def test_imagenet__Gradient():
def method(model):
return Gradient(model)
dryrun.test_analyzer(method, "imagenet.*")
def test_precommit__BaselineGradient_pp_square():
def method(model):
return BaselineGradient(model, postprocess="square")
dryrun.test_analyzer(method, "mnist.*")
def test_fast__Gradient():
def method(model):
return Gradient(model)
dryrun.test_analyzer(method, "trivia.*:mnist.log_reg")
def test_imagenet__SmoothGrad():
def method(model):
return SmoothGrad(model, augment_by_n=2)
dryrun.test_analyzer(method, "imagenet.*")
def test_fast__BaselineGradient_pp_square():
def method(model):
return BaselineGradient(model, postprocess="square")
dryrun.test_analyzer(method, "trivia.*:mnist.log_reg")
def test_precommit__BoundedDeepTaylor():
def method(model):
return BoundedDeepTaylor(model, low=-1, high=1)
dryrun.test_analyzer(method, "mnist.*")
def test_fast__GuidedBackprop():
def method(model):
return GuidedBackprop(model)
dryrun.test_analyzer(method, "trivia.*:mnist.log_reg")
def test_fast__DeepTaylor():
def method(model):
return DeepTaylor(model)
dryrun.test_analyzer(method, "trivia.*:mnist.log_reg")
def test_imagenet__DeepTaylor():
def method(model):
return DeepTaylor(model)
dryrun.test_analyzer(method, "imagenet.*")
def test_fast__IntegratedGradients():
def method(model):
return IntegratedGradients(model)
dryrun.test_analyzer(method, "trivia.*:mnist.log_reg")
def test_imagenet__GuidedBackprop():
def method(model):
return GuidedBackprop(model)
dryrun.test_analyzer(method, "imagenet.*")
def test_precommit__GuidedBackprop():
def method(model):
return GuidedBackprop(model)
dryrun.test_analyzer(method, "mnist.*")
def test_precommit__Gradient_pp_None():
def method(model):
return Gradient(model, postprocess=None)
dryrun.test_analyzer(method, "mnist.*")
def test_precommit__Deconvnet():
def method(model):
return Deconvnet(model)
dryrun.test_analyzer(method, "mnist.*")
def test_precommit__IntegratedGradients():
def method(model):
return IntegratedGradients(model)
dryrun.test_analyzer(method, "mnist.*")
def test_imagenet__Deconvnet():
def method(model):
return Deconvnet(model)
dryrun.test_analyzer(method, "imagenet.*")
def test_precommit__DeepTaylor():
def method(model):
return DeepTaylor(model)
dryrun.test_analyzer(method, "mnist.*")
def test_imagenet__IntegratedGradients():
def method(model):
return IntegratedGradients(model, steps=2)
dryrun.test_analyzer(method, "imagenet.*")
def test_fast__BoundedDeepTaylor():
def method(model):
return BoundedDeepTaylor(model, low=-1, high=1)
dryrun.test_analyzer(method, "trivia.*:mnist.log_reg")
def test_fast__SmoothGrad():
def method(model):
return SmoothGrad(model)
dryrun.test_analyzer(method, "trivia.*:mnist.log_reg")
def test_imagenet__BoundedDeepTaylor():
def method(model):
return BoundedDeepTaylor(model, low=-1, high=1)
dryrun.test_analyzer(method, "imagenet.*")
def test_precommit__SmoothGrad():
def method(model):
return SmoothGrad(model)
dryrun.test_analyzer(method, "mnist.*")
