Update code structure.

1. Fixed linter errors;
2. Added ,travis;
3. Added Sum layer;
4. Minor tests clean up;
5. Changed imports to package-relatives.

.travis.yml

@@ -0,0 +1,27 @@
+group: travis_latest
+language: python
+cache: pip
+python:
+    - 2.7
+    - 3.6
+    #- nightly
+    #- pypy
+    #- pypy3
+matrix:
+    allow_failures:
+        - python: nightly
+        - python: pypy
+        - python: pypy3
+install:
+    #- pip install -r requirements.txt
+    - pip install flake8  # pytest  # add another testing frameworks later
+before_script:
+    # stop the build if there are Python syntax errors or undefined names
+    - flake8 . --count --select=E901,E999,F821,F822,F823 --show-source --statistics
+    # exit-zero treats all errors as warnings.  The GitHub editor is 127 chars wide
+    - flake8 . --count --exit-zero --max-complexity=32 --max-line-length=127 --statistics
+script:
+    - true  # pytest --capture=sys  # add other tests here
+notifications:
+    on_success: change
+    on_failure: change  # `always` will be the setting once code changes slow down

pytorch2keras/converter.py

@@ -9,7 +9,7 @@
 import contextlib
 from torch.jit import _unique_state_dict
 
-from layers import AVAILABLE_CONVERTERS
+from .layers import AVAILABLE_CONVERTERS
 
 
 @contextlib.contextmanager

pytorch2keras/layers.py

@@ -436,6 +436,29 @@ def convert_elementwise_sub(
     layers[scope_name] = sub([model0, model1])
 
 
+def convert_sum(
+    params, w_name, scope_name, inputs, layers, weights
+):
+    """
+    Convert sum.
+
+    Args:
+        params: dictionary with layer parameters
+        w_name: name prefix in state_dict
+        scope_name: pytorch scope name
+        inputs: pytorch node inputs
+        layers: dictionary with keras tensors
+        weights: pytorch state_dict
+    """
+    print('Converting Sum ...')
+
+    def target_layer(x):
+        return keras.backend.sum(x)
+
+    lambda_layer = keras.layers.Lambda(target_layer)
+    layers[scope_name] = lambda_layer(layers[inputs[0]])
+
+
 def convert_concat(params, w_name, scope_name, inputs, layers, weights):
     """
     Convert concatenation.
@@ -469,6 +492,7 @@ def convert_relu(params, w_name, scope_name, inputs, layers, weights):
     """
     print('Converting relu ...')
 
+    print(w_name, scope_name)
     tf_name = w_name + str(random.random())
     relu = keras.layers.Activation('relu', name=tf_name)
     layers[scope_name] = relu(layers[inputs[0]])
@@ -570,7 +594,6 @@ def convert_selu(params, w_name, scope_name, inputs, layers, weights):
     layers[scope_name] = selu(layers[inputs[0]])
 
 
-
 def convert_transpose(params, w_name, scope_name, inputs, layers, weights):
     """
     Convert transpose layer.
@@ -705,7 +728,9 @@ def convert_reduce_sum(params, w_name, scope_name, inputs, layers, weights):
 
     keepdims = params['keepdims'] > 0
     axis = np.array(params['axes'])
-    target_layer = lambda x: keras.backend.sum(x, keepdims=keepdims, axis=axis)
+
+    def target_layer(x, keepdims=keepdims, axis=axis):
+        return keras.backend.sum(x, keepdims=keepdims, axis=axis)
 
     lambda_layer = keras.layers.Lambda(target_layer)
     layers[scope_name] = lambda_layer(layers[inputs[0]])
@@ -725,7 +750,9 @@ def convert_constant(params, w_name, scope_name, inputs, layers, weights):
     """
     print('Converting constant ...')
 
-    target_layer = lambda x: keras.backend.constant(np.float32(params['value']))
+    def target_layer(params=params):
+        return keras.backend.constant(np.float32(params['value']))
+
     lambda_layer = keras.layers.Lambda(target_layer)
     layers[scope_name] = lambda_layer(layers[inputs[0]])
 
@@ -782,7 +809,7 @@ def convert_padding(params, w_name, scope_name, inputs, layers, weights):
     padding_name = tf_name + '_pad'
     padding_layer = keras.layers.ZeroPadding2D(
         padding=((params['pads'][2], params['pads'][6]), (params['pads'][3], params['pads'][7])),
-        name=tf_name
+        name=padding_name
     )
 
     layers[scope_name] = padding_layer(layers[inputs[0]])
@@ -801,6 +828,7 @@ def convert_padding(params, w_name, scope_name, inputs, layers, weights):
     'onnx::Add': convert_elementwise_add,
     'onnx::Mul': convert_elementwise_mul,
     'onnx::Sub': convert_elementwise_sub,
+    'onnx::Sum': convert_sum,
     'onnx::Concat': convert_concat,
     'onnx::Relu': convert_relu,
     'onnx::LeakyRelu': convert_lrelu,

setup.py

@@ -1,20 +1,21 @@
 from setuptools import setup, find_packages
-from setuptools.command.develop import develop
-from setuptools.command.install import install
 
 
-try: # for pip >= 10
+try:  # for pip >= 10
     from pip._internal.req import parse_requirements
-except ImportError: # for pip <= 9.0.3
+except ImportError:  # for pip <= 9.0.3
     from pip.req import parse_requirements
 
+
 # parse_requirements() returns generator of pip.req.InstallRequirement objects
 install_reqs = parse_requirements('requirements.txt', session='null')
 
+
 # reqs is a list of requirement
 # e.g. ['django==1.5.1', 'mezzanine==1.4.6']
 reqs = [str(ir.req) for ir in install_reqs]
 
+
 setup(name='pytorch2keras',
       version='0.1',
       description='The model convertor',

tests/alexnet.py

@@ -1,14 +1,8 @@
-import keras  # work around segfault
-import sys
 import numpy as np
-
 import torch
-import torchvision
 from torch.autograd import Variable
-
-sys.path.append('../pytorch2keras')
-from converter import pytorch_to_keras
-
+from pytorch2keras.converter import pytorch_to_keras
+import torchvision
 
 if __name__ == '__main__':
     max_error = 0

tests/avg_pool.py

@@ -1,13 +1,8 @@
-import keras  # work around segfault
-import sys
 import numpy as np
-
 import torch
 import torch.nn as nn
 from torch.autograd import Variable
-
-sys.path.append('../pytorch2keras')
-from converter import pytorch_to_keras
+from pytorch2keras.converter import pytorch_to_keras
 
 
 class AvgPool(nn.Module):

tests/bn.py

@@ -1,13 +1,8 @@
-import keras  # work around segfault
-import sys
 import numpy as np
-
 import torch
 import torch.nn as nn
 from torch.autograd import Variable
-
-sys.path.append('../pytorch2keras')
-from converter import pytorch_to_keras
+from pytorch2keras.converter import pytorch_to_keras
 
 
 class TestConv2d(nn.Module):

tests/concat_many.py

@@ -1,13 +1,8 @@
-import keras  # work around segfault
-import sys
 import numpy as np
-
 import torch
 import torch.nn as nn
 from torch.autograd import Variable
-
-sys.path.append('../pytorch2keras')
-from converter import pytorch_to_keras
+from pytorch2keras.converter import pytorch_to_keras
 
 
 class TestConcatMany(nn.Module):

tests/const.py

@@ -1,13 +1,8 @@
-import keras  # work around segfault
-import sys
 import numpy as np
-
 import torch
 import torch.nn as nn
 from torch.autograd import Variable
-
-sys.path.append('../pytorch2keras')
-from converter import pytorch_to_keras
+from pytorch2keras.converter import pytorch_to_keras
 
 
 class TestConst(nn.Module):

tests/conv2d.py

@@ -1,13 +1,8 @@
-import keras  # work around segfault
-import sys
 import numpy as np
-
 import torch
 import torch.nn as nn
 from torch.autograd import Variable
-
-sys.path.append('../pytorch2keras')
-from converter import pytorch_to_keras
+from pytorch2keras.converter import pytorch_to_keras
 
 
 class TestConv2d(nn.Module):

tests/conv2d_channels_last.py

@@ -1,13 +1,8 @@
-import keras  # work around segfault
-import sys
 import numpy as np
-
 import torch
 import torch.nn as nn
 from torch.autograd import Variable
-
-sys.path.append('../pytorch2keras')
-from converter import pytorch_to_keras
+from pytorch2keras.converter import pytorch_to_keras
 
 
 class TestConv2d(nn.Module):

tests/conv2d_dilation.py

@@ -1,13 +1,8 @@
-import keras  # work around segfault
-import sys
 import numpy as np
-
 import torch
 import torch.nn as nn
 from torch.autograd import Variable
-
-sys.path.append('../pytorch2keras')
-from converter import pytorch_to_keras
+from pytorch2keras.converter import pytorch_to_keras
 
 
 class TestConv2d(nn.Module):

tests/convtranspose2d.py

@@ -1,22 +1,17 @@
-import keras  # work around segfault
-import sys
 import numpy as np
-
 import torch
 import torch.nn as nn
 from torch.autograd import Variable
-
-sys.path.append('../pytorch2keras')
-from converter import pytorch_to_keras
+from pytorch2keras.converter import pytorch_to_keras
 
 
 class TestConvTranspose2d(nn.Module):
     """Module for ConvTranspose2d conversion testing
     """
 
-    def __init__(self, inp=10, out=16, kernel_size=3, bias=True):
+    def __init__(self, inp=10, out=16, kernel_size=3, padding=1, bias=True):
         super(TestConvTranspose2d, self).__init__()
-        self.conv2d = nn.ConvTranspose2d(inp, out, kernel_size=kernel_size, bias=bias)
+        self.conv2d = nn.ConvTranspose2d(inp, out, kernel_size=kernel_size, bias=bias, stride=padding)
 
     def forward(self, x):
         x = self.conv2d(x)
@@ -30,7 +25,7 @@ def forward(self, x):
         inp = np.random.randint(kernel_size + 1, 100)
         out = np.random.randint(1, 100)
 
-        model = TestConvTranspose2d(inp, out, kernel_size, inp % 2)
+        model = TestConvTranspose2d(inp, out, kernel_size, 2, inp % 3)
 
         input_np = np.random.uniform(0, 1, (1, inp, inp, inp))
         input_var = Variable(torch.FloatTensor(input_np))

tests/dense.py

@@ -1,13 +1,8 @@
-import keras  # work around segfault
-import sys
 import numpy as np
-
 import torch
 import torch.nn as nn
 from torch.autograd import Variable
-
-sys.path.append('../pytorch2keras')
-from converter import pytorch_to_keras
+from pytorch2keras.converter import pytorch_to_keras
 
 
 class TestDense(nn.Module):

tests/densenet.py

@@ -1,15 +1,9 @@
-import keras  # work around segfault
-import sys
 import numpy as np
-import math
-
 import torch
-import torchvision
-import torch.nn as nn
 from torch.autograd import Variable
+from pytorch2keras.converter import pytorch_to_keras
+import torchvision
 
-sys.path.append('../pytorch2keras')
-from converter import pytorch_to_keras
 
 if __name__ == '__main__':
     max_error = 0

tests/droupout.py

@@ -1,13 +1,8 @@
-import keras  # work around segfault
-import sys
 import numpy as np
-
 import torch
 import torch.nn as nn
 from torch.autograd import Variable
-
-sys.path.append('../pytorch2keras')
-from converter import pytorch_to_keras
+from pytorch2keras.converter import pytorch_to_keras
 
 
 class TestDropout(nn.Module):

tests/embedding.py

@@ -1,13 +1,8 @@
-import keras  # work around segfault
-import sys
 import numpy as np
-
 import torch
 import torch.nn as nn
 from torch.autograd import Variable
-
-sys.path.append('../pytorch2keras')
-from converter import pytorch_to_keras
+from pytorch2keras.converter import pytorch_to_keras
 
 
 class TestEmbedding(nn.Module):
@@ -16,7 +11,7 @@ def __init__(self, input_size):
         self.embedd = nn.Embedding(input_size, 100)
 
     def forward(self, input):
-        return self.embedd(input).sum(dim=0)
+        return self.embedd(input)
 
 
 if __name__ == '__main__':