Adds compatibility with python 2.7 (#16)

closes #16 

* Fixed Python 2.7 compatibility

* Removed residual debug print

* Fixed PEP8 compatibility

* added python 2.7 to travis config

* install python 2 requirements if python2.7

* addressing requested changes

* fixed bug in travis config

.travis.yml

@@ -2,17 +2,19 @@ sudo: false
 dist: trusty
 language: python
 python:
+    - 2.7
     - 3.5
     - 3.6
 before_install:
     - pip install -U pip
 install:
     - travis_wait travis_retry pip install --upgrade numpy
     - travis_wait travis_retry pip install --upgrade scipy
-    - travis_wait travis_retry pip install -r requirements-dev.txt
+    - if [[ $TRAVIS_PYTHON_VERSION == 2.7 ]]; then travis_wait travis_retry pip install -r requirements-dev-py2.txt; else travis_wait travis_retry pip install -r requirements-dev.txt; fi
     - travis_wait travis_retry pip install --upgrade tensorflow
     - travis_wait travis_retry pip install --upgrade theano
     - travis_wait travis_retry pip install --upgrade https://github.com/Lasagne/Lasagne/archive/master.zip
+    - if [[ $TRAVIS_PYTHON_VERSION == 2.7 ]]; then pip install http://download.pytorch.org/whl/cu75/torch-0.1.12.post2-cp27-none-linux_x86_64.whl; fi
     - if [[ $TRAVIS_PYTHON_VERSION == 3.5 ]]; then pip install http://download.pytorch.org/whl/cu75/torch-0.1.12.post2-cp35-cp35m-linux_x86_64.whl; fi
     - if [[ $TRAVIS_PYTHON_VERSION == 3.6 ]]; then pip install http://download.pytorch.org/whl/cu75/torch-0.1.12.post2-cp36-cp36m-linux_x86_64.whl; fi
     - travis_wait travis_retry pip install --upgrade keras

foolbox/adversarial.py

@@ -37,7 +37,6 @@ def __init__(
             criterion,
             original_image,
             original_class,
-            *,
             distance=MSE):
 
         self.__model = model
@@ -154,7 +153,7 @@ def in_bounds(self, input_):
         min_, max_ = self.bounds()
         return min_ <= input_.min() and input_.max() <= max_
 
-    def channel_axis(self, *, batch):
+    def channel_axis(self, batch):
         """Interface to model.channel_axis for attacks.
 
         Parameters
@@ -202,7 +201,7 @@ def predictions(self, image, strict=True):
         assert predictions.ndim == 1
         return predictions, is_adversarial
 
-    def batch_predictions(self, images, *, increasing=False, strict=True):
+    def batch_predictions(self, images, increasing=False, strict=True):
         """Interface to model.batch_predictions for attacks.
 
         Parameters

foolbox/attacks/base.py

@@ -1,5 +1,12 @@
 import logging
-from abc import ABC, abstractmethod
+import sys
+import abc
+abstractmethod = abc.abstractmethod
+
+if sys.version_info >= (3, 4):
+    ABC = abc.ABC
+else:
+    ABC = abc.ABCMeta('ABC', (), {})
 
 from ..adversarial import Adversarial
 from ..criteria import Misclassification
@@ -36,7 +43,6 @@ def __call__(
             self,
             image,
             label=None,
-            *,
             unpack=True,
             **kwargs):
 

foolbox/attacks/gradient.py

@@ -1,3 +1,4 @@
+from __future__ import division
 import numpy as np
 from collections import Iterable
 

foolbox/attacks/gradientsign.py

@@ -1,3 +1,4 @@
+from __future__ import division
 import numpy as np
 from collections import Iterable
 

foolbox/attacks/lbfgs.py

@@ -1,3 +1,4 @@
+from __future__ import division
 import random
 import logging
 
@@ -30,9 +31,14 @@ class LBFGSAttack(Attack):
 
     """
 
-    def __init__(self, *args, approximate_gradient=False, **kwargs):
-        super().__init__(*args, **kwargs)
-        self._approximate_gradient = approximate_gradient
+    def __init__(self, *args, **kwargs):
+        if 'approximate_gradient' in kwargs:
+            self._approximate_gradient = kwargs['approximate_gradient']
+            del kwargs['approximate_gradient']
+            super(LBFGSAttack, self).__init__(*args, **kwargs)
+        else:
+            self._approximate_gradient = False
+            super(LBFGSAttack, self).__init__(*args, **kwargs)
 
     def name(self):
         prefix = 'Approximate' if self._approximate_gradient else ''
@@ -212,4 +218,5 @@ class ApproximateLBFGSAttack(LBFGSAttack):
 
     def __init__(self, *args, **kwargs):
         assert 'approximate_gradient' not in kwargs
-        super().__init__(*args, approximate_gradient=True, **kwargs)
+        kwargs['approximate_gradient'] = True
+        super(ApproximateLBFGSAttack, self).__init__(*args, **kwargs)

foolbox/attacks/localsearch.py

@@ -1,3 +1,4 @@
+from __future__ import division
 import numpy as np
 
 from .base import Attack
@@ -117,7 +118,6 @@ def cyclic(r, Ibxy):
         PxPy = random_locations()
 
         for _ in range(R):
-
             # Computing the function g using the neighborhood
             L = [pert(Ii, p, x, y) for x, y in PxPy]
 
@@ -149,8 +149,8 @@ def score(It):
             # Update a neighborhood of pixel locations for the next round
             PxPy = [
                 (x, y)
-                for a, b in PxPy_star
-                for x in range(a - d, a + d + 1)
-                for y in range(b - d, b + d + 1)]
+                for _a, _b in PxPy_star
+                for x in range(_a - d, _a + d + 1)
+                for y in range(_b - d, _b + d + 1)]
             PxPy = [(x, y) for x, y in PxPy if 0 <= x < w and 0 <= y < h]
             PxPy = np.array(PxPy)

foolbox/attacks/precomputed.py

@@ -15,7 +15,7 @@ class PrecomputedImagesAttack(Attack):
     """
 
     def __init__(self, input_images, output_images, *args, **kwargs):
-        super().__init__(*args, **kwargs)
+        super(PrecomputedImagesAttack, self).__init__(*args, **kwargs)
 
         assert input_images.shape == output_images.shape
 

foolbox/attacks/saltandpepper.py

@@ -9,7 +9,7 @@ class SaltAndPepperNoiseAttack(Attack):
 
     """
 
-    def _apply(self, a, *, epsilons=100, repetitions=10):
+    def _apply(self, a, epsilons=100, repetitions=10):
         image = a.original_image
         min_, max_ = a.bounds()
         axis = a.channel_axis(batch=False)

foolbox/attacks/slsqp.py

@@ -12,7 +12,7 @@ class SLSQPAttack(Attack):
     # is differentiable) and use this to provide constraint gradients
 
     def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
+        super(SLSQPAttack, self).__init__(*args, **kwargs)
         self.last_result = None
 
     def _apply(self, a):

foolbox/criteria.py

@@ -43,8 +43,14 @@
 >>> criterion5 = criterion2 & criterion3
 
 """
+import sys
+import abc
+abstractmethod = abc.abstractmethod
 
-from abc import ABC, abstractmethod
+if sys.version_info >= (3, 4):
+    ABC = abc.ABC
+else:
+    ABC = abc.ABCMeta('ABC', (), {})
 
 import numpy as np
 
@@ -127,7 +133,7 @@ class CombinedCriteria(Criterion):
     """
 
     def __init__(self, *criteria):
-        super().__init__()
+        super(CombinedCriteria, self).__init__()
         self._criteria = criteria
 
     def name(self):
@@ -203,8 +209,8 @@ class TopKMisclassification(Criterion):
 
     """
 
-    def __init__(self, *, k):
-        super().__init__()
+    def __init__(self, k):
+        super(TopKMisclassification, self).__init__()
         self.k = k
 
     def name(self):
@@ -232,7 +238,7 @@ class TargetClass(Criterion):
     """
 
     def __init__(self, target_class):
-        super().__init__()
+        super(TargetClass, self).__init__()
         self._target_class = target_class
 
     def target_class(self):
@@ -265,7 +271,7 @@ class OriginalClassProbability(Criterion):
     """
 
     def __init__(self, p):
-        super().__init__()
+        super(OriginalClassProbability, self).__init__()
         assert 0 <= p <= 1
         self.p = p
 
@@ -299,8 +305,8 @@ class TargetClassProbability(Criterion):
 
     """
 
-    def __init__(self, target_class, *, p):
-        super().__init__()
+    def __init__(self, target_class, p):
+        super(TargetClassProbability, self).__init__()
         self._target_class = target_class
         assert 0 <= p <= 1
         self.p = p

foolbox/distances.py

@@ -30,8 +30,16 @@
    Distance
 
 """
+from __future__ import division
+import sys
+import abc
+abstractmethod = abc.abstractmethod
+
+if sys.version_info >= (3, 4):
+    ABC = abc.ABC
+else:
+    ABC = abc.ABCMeta('ABC', (), {})
 
-from abc import ABC, abstractmethod
 import functools
 import numpy as np
 from numbers import Number
@@ -50,7 +58,6 @@ def __init__(
             self,
             reference=None,
             other=None,
-            *,
             bounds=None,
             value=None):
 
@@ -99,12 +106,12 @@ def __repr__(self):
     def __eq__(self, other):
         if other.__class__ != self.__class__:
             raise TypeError('Comparisons are only possible between the same distance types.')  # noqa: E501
-        return self.value.__eq__(other.value)
+        return self.value == other.value
 
     def __lt__(self, other):
         if other.__class__ != self.__class__:
             raise TypeError('Comparisons are only possible between the same distance types.')  # noqa: E501
-        return self.value.__lt__(other.value)
+        return self.value < other.value
 
 
 class MeanSquaredDistance(Distance):

foolbox/models/base.py

@@ -1,5 +1,12 @@
 import numpy as np
-from abc import ABC, abstractmethod
+import sys
+import abc
+abstractmethod = abc.abstractmethod
+
+if sys.version_info >= (3, 4):
+    ABC = abc.ABC
+else:
+    ABC = abc.ABCMeta('ABC', (), {})
 
 
 class Model(ABC):
@@ -22,7 +29,7 @@ class Model(ABC):
 
     """
 
-    def __init__(self, *, bounds, channel_axis):
+    def __init__(self, bounds, channel_axis):
         assert len(bounds) == 2
         self._bounds = bounds
         assert channel_axis in [0, 1, 2, 3]

foolbox/models/keras.py

@@ -1,5 +1,5 @@
+from __future__ import absolute_import
 import warnings
-
 import numpy as np
 
 from .base import DifferentiableModel
@@ -28,13 +28,13 @@ class KerasModel(DifferentiableModel):
     def __init__(
             self,
             model,
-            *,
             bounds,
             channel_axis=3,
             predicts='probabilities',
             preprocess_fn=None):
 
-        super().__init__(bounds=bounds, channel_axis=channel_axis)
+        super(KerasModel, self).__init__(bounds=bounds,
+                                         channel_axis=channel_axis)
 
         from keras import backend as K
 

foolbox/models/lasagne.py

@@ -1,3 +1,4 @@
+from __future__ import absolute_import
 import numpy as np
 
 
@@ -25,11 +26,11 @@ def __init__(
             self,
             input_layer,
             logits_layer,
-            *,
             bounds,
             channel_axis=1):
 
-        super().__init__(bounds=bounds, channel_axis=channel_axis)
+        super(LasagneModel, self).__init__(bounds=bounds,
+                                           channel_axis=channel_axis)
 
         # delay import until class is instantiated
         import theano as th

foolbox/models/pytorch.py

@@ -25,13 +25,13 @@ class PyTorchModel(DifferentiableModel):
     def __init__(
             self,
             model,
-            *,
             bounds,
             num_classes,
             channel_axis=1,
             cuda=True):
 
-        super().__init__(bounds=bounds, channel_axis=channel_axis)
+        super(PyTorchModel, self).__init__(bounds=bounds,
+                                           channel_axis=channel_axis)
 
         self._num_classes = num_classes
         self._model = model

foolbox/models/tensorflow.py

@@ -1,5 +1,5 @@
+from __future__ import absolute_import
 import numpy as np
-
 from .base import DifferentiableModel
 
 
@@ -24,11 +24,11 @@ def __init__(
             self,
             images,
             logits,
-            *,
             bounds,
             channel_axis=3):
 
-        super().__init__(bounds=bounds, channel_axis=channel_axis)
+        super(TensorFlowModel, self).__init__(bounds=bounds,
+                                              channel_axis=channel_axis)
 
         # delay import until class is instantiated
         import tensorflow as tf

foolbox/models/theano.py

@@ -1,6 +1,5 @@
+from __future__ import absolute_import
 import numpy as np
-
-
 from .base import DifferentiableModel
 
 
@@ -27,12 +26,12 @@ def __init__(
             self,
             images,
             logits,
-            *,
             bounds,
             num_classes,
             channel_axis=1):
 
-        super().__init__(bounds=bounds, channel_axis=channel_axis)
+        super(TheanoModel, self).__init__(bounds=bounds,
+                                          channel_axis=channel_axis)
 
         self._num_classes = num_classes