Merge pull request #909 from f0k/int_types

Properly handle integral types other than plain int

docs/modules/utils.rst

@@ -3,6 +3,8 @@
 
 .. automodule:: lasagne.utils
 
+.. autodata:: int_types
+ :annotation: = (numbers.Integral, np.integer)
 .. autofunction:: floatX
 .. autofunction:: shared_empty
 .. autofunction:: as_theano_expression

lasagne/layers/conv.py

@@ -2,7 +2,7 @@
 
 from .. import init
 from .. import nonlinearities
-from ..utils import as_tuple, inspect_kwargs
+from ..utils import as_tuple, int_types, inspect_kwargs
 from ..theano_extensions import conv
 
 from .base import Layer
@@ -75,7 +75,7 @@ def conv_output_length(input_length, filter_size, stride, pad=0):
  output_length = input_length + filter_size - 1
  elif pad == 'same':
  output_length = input_length
- elif isinstance(pad, int):
+ elif isinstance(pad, int_types):
  output_length = input_length + 2 * pad - filter_size + 1
  else:
  raise ValueError('Invalid pad: {0}'.format(pad))
@@ -151,7 +151,7 @@ def conv_input_length(output_length, filter_size, stride, pad=0):
  pad = filter_size - 1
  elif pad == 'same':
  pad = filter_size // 2
- if not isinstance(pad, int):
+ if not isinstance(pad, int_types):
  raise ValueError('Invalid pad: {0}'.format(pad))
  return (output_length - 1) * stride - 2 * pad + filter_size
 
@@ -288,9 +288,9 @@ def __init__(self, incoming, num_filters, filter_size, stride=1, pad=0,
  (n, self.input_shape, n+2, n))
  self.n = n
  self.num_filters = num_filters
- self.filter_size = as_tuple(filter_size, n, int)
+ self.filter_size = as_tuple(filter_size, n, int_types)
  self.flip_filters = flip_filters
- self.stride = as_tuple(stride, n, int)
+ self.stride = as_tuple(stride, n, int_types)
  self.untie_biases = untie_biases
 
  if pad == 'same':
@@ -302,7 +302,7 @@ def __init__(self, incoming, num_filters, filter_size, stride=1, pad=0,
  elif pad in ('full', 'same'):
  self.pad = pad
  else:
- self.pad = as_tuple(pad, n, int)
+ self.pad = as_tuple(pad, n, int_types)
 
  if (num_groups <= 0 or
  self.num_filters % num_groups != 0 or
@@ -934,7 +934,7 @@ def __init__(self, incoming, num_filters, filter_size, stride=(1, 1),
  # output_size must be set before calling the super constructor
  if (not isinstance(output_size, T.Variable) and
  output_size is not None):
- output_size = as_tuple(output_size, 2, int)
+ output_size = as_tuple(output_size, 2, int_types)
  self.output_size = output_size
  super(TransposedConv2DLayer, self).__init__(
  incoming, num_filters, filter_size, stride, crop, untie_biases,
@@ -1127,7 +1127,7 @@ def __init__(self, incoming, num_filters, filter_size,
  # output_size must be set before calling the super constructor
  if (not isinstance(output_size, T.Variable) and
  output_size is not None):
- output_size = as_tuple(output_size, 3, int)
+ output_size = as_tuple(output_size, 3, int_types)
  self.output_size = output_size
  BaseConvLayer.__init__(self, incoming, num_filters, filter_size,
  stride, crop, untie_biases, W, b,
@@ -1286,7 +1286,7 @@ def __init__(self, incoming, num_filters, filter_size, dilation=(1, 1),
  W=init.GlorotUniform(), b=init.Constant(0.),
  nonlinearity=nonlinearities.rectify, flip_filters=False,
  **kwargs):
- self.dilation = as_tuple(dilation, 2, int)
+ self.dilation = as_tuple(dilation, 2, int_types)
  super(DilatedConv2DLayer, self).__init__(
  incoming, num_filters, filter_size, 1, pad,
  untie_biases, W, b, nonlinearity, flip_filters, n=2, **kwargs)

lasagne/layers/normalization.py

@@ -41,6 +41,7 @@
 
 from .. import init
 from .. import nonlinearities
+from ..utils import int_types
 
 from .base import Layer
 
@@ -235,7 +236,7 @@ def __init__(self, incoming, axes='auto', epsilon=1e-4, alpha=0.1,
  if axes == 'auto':
  # default: normalize over all but the second axis
  axes = (0,) + tuple(range(2, len(self.input_shape)))
- elif isinstance(axes, int):
+ elif isinstance(axes, int_types):
  axes = (axes,)
  self.axes = axes
 

lasagne/layers/shape.py

@@ -2,6 +2,7 @@
 import theano.tensor as T
 
 from ..theano_extensions import padding
+from ..utils import int_types
 
 from .base import Layer
 
@@ -105,11 +106,11 @@ def __init__(self, incoming, shape, **kwargs):
  super(ReshapeLayer, self).__init__(incoming, **kwargs)
  shape = tuple(shape)
  for s in shape:
- if isinstance(s, int):
+ if isinstance(s, int_types):
  if s == 0 or s < - 1:
  raise ValueError("`shape` integers must be positive or -1")
  elif isinstance(s, list):
- if len(s) != 1 or not isinstance(s[0], int) or s[0] < 0:
+ if len(s) != 1 or not isinstance(s[0], int_types) or s[0] < 0:
  raise ValueError("`shape` input references must be "
  "single-element lists of int >= 0")
  elif isinstance(s, T.TensorVariable):
@@ -232,7 +233,7 @@ def __init__(self, incoming, pattern, **kwargs):
  # Sanity check the pattern
  used_dims = set()
  for p in pattern:
- if isinstance(p, int):
+ if isinstance(p, int_types):
  # Dimension p
  if p in used_dims:
  raise ValueError("pattern contains dimension {0} more "
@@ -256,7 +257,7 @@ def get_output_shape_for(self, input_shape):
  output_shape = []
  dims_used = [False] * len(input_shape)
  for p in self.pattern:
- if isinstance(p, int):
+ if isinstance(p, int_types):
  if p < 0 or p >= len(input_shape):
  raise ValueError("pattern contains {0}, but input shape "
  "has {1} dimensions "
@@ -321,7 +322,7 @@ def __init__(self, incoming, width, val=0, batch_ndim=2, **kwargs):
  def get_output_shape_for(self, input_shape):
  output_shape = list(input_shape)
 
- if isinstance(self.width, int):
+ if isinstance(self.width, int_types):
  widths = [self.width] * (len(input_shape) - self.batch_ndim)
  else:
  widths = self.width
@@ -381,7 +382,7 @@ def __init__(self, incoming, indices, axis=-1, **kwargs):
 
  def get_output_shape_for(self, input_shape):
  output_shape = list(input_shape)
- if isinstance(self.slice, int):
+ if isinstance(self.slice, int_types):
  del output_shape[self.axis]
  elif input_shape[self.axis] is not None:
  output_shape[self.axis] = len(

lasagne/layers/special.py

@@ -4,7 +4,7 @@
 
 from .. import init
 from .. import nonlinearities
-from ..utils import as_tuple, floatX
+from ..utils import as_tuple, floatX, int_types
 from ..random import get_rng
 from .base import Layer, MergeLayer
 from theano.sandbox.rng_mrg import MRG_RandomStreams as RandomStreams
@@ -94,7 +94,7 @@ def __init__(self, incoming, b=init.Constant(0), shared_axes='auto',
  if shared_axes == 'auto':
  # default: share biases over all but the second axis
  shared_axes = (0,) + tuple(range(2, len(self.input_shape)))
- elif isinstance(shared_axes, int):
+ elif isinstance(shared_axes, int_types):
  shared_axes = (shared_axes,)
  self.shared_axes = shared_axes
 
@@ -161,7 +161,7 @@ def __init__(self, incoming, scales=init.Constant(1), shared_axes='auto',
  if shared_axes == 'auto':
  # default: share scales over all but the second axis
  shared_axes = (0,) + tuple(range(2, len(self.input_shape)))
- elif isinstance(shared_axes, int):
+ elif isinstance(shared_axes, int_types):
  shared_axes = (shared_axes,)
  self.shared_axes = shared_axes
 
@@ -1007,7 +1007,7 @@ def __init__(self, incoming, alpha=init.Constant(0.25), shared_axes='auto',
  self.shared_axes = (0,) + tuple(range(2, len(self.input_shape)))
  elif shared_axes == 'all':
  self.shared_axes = tuple(range(len(self.input_shape)))
- elif isinstance(shared_axes, int):
+ elif isinstance(shared_axes, int_types):
  self.shared_axes = (shared_axes,)
  else:
  self.shared_axes = shared_axes
@@ -1120,7 +1120,7 @@ def __init__(self, incoming, lower=0.3, upper=0.8, shared_axes='auto',
  self.shared_axes = (0,) + tuple(range(2, len(self.input_shape)))
  elif shared_axes == 'all':
  self.shared_axes = tuple(range(len(self.input_shape)))
- elif isinstance(shared_axes, int):
+ elif isinstance(shared_axes, int_types):
  self.shared_axes = (shared_axes,)
  else:
  self.shared_axes = shared_axes

lasagne/tests/layers/test_conv.py

@@ -399,6 +399,11 @@ def test_fail_on_mismatching_groups(self):
  BaseConvLayer((2, 3, 4), 1, 3, num_groups=-3)
  assert "must be positive" in exc.value.args[0]
 
+ def test_integer_types(self):
+ from lasagne.layers.conv import BaseConvLayer
+ BaseConvLayer((2, 3, 4), np.int64(1), np.int64(3))
+ BaseConvLayer((2, 3, 4, 5), 1, np.empty((3, 3)).shape)
+
 
 class TestConv1DLayer:
 

lasagne/tests/test_utils.py

@@ -5,6 +5,21 @@
 import theano.tensor as T
 
 
+def test_int_types():
+ from lasagne.utils import int_types
+ assert isinstance(42, int_types)
+ assert isinstance(np.int8(42), int_types)
+ assert isinstance(np.int16(42), int_types)
+ assert isinstance(np.int32(42), int_types)
+ assert isinstance(np.int64(42), int_types)
+ assert isinstance(np.empty(42).shape[0], int_types)
+ assert isinstance(np.prod(np.empty(42).shape), int_types)
+ try:
+ assert isinstance(long(42), int_types)
+ except NameError:
+ pass
+
+
 def test_shared_empty():
  from lasagne.utils import shared_empty
 
@@ -45,11 +60,19 @@ def test_one_hot():
 
 
 def test_as_tuple_fails():
- from lasagne.utils import as_tuple
- with pytest.raises(ValueError):
+ from lasagne.utils import as_tuple, int_types
+ with pytest.raises(ValueError) as exc:
  as_tuple([1, 2, 3], 4)
- with pytest.raises(TypeError):
+ assert "length 4" in exc.value.args[0]
+ with pytest.raises(TypeError) as exc:
  as_tuple('asdf', 4, int)
+ assert "of int," in exc.value.args[0]
+ with pytest.raises(TypeError) as exc:
+ as_tuple('asdf', 4, (int, float))
+ assert "of int or float," in exc.value.args[0]
+ with pytest.raises(TypeError) as exc:
+ as_tuple('asdf', 4, int_types)
+ assert "of int," in exc.value.args[0]
 
 
 def test_inspect_kwargs():

lasagne/theano_extensions/conv.py

@@ -3,9 +3,10 @@
 """
 
 import numpy as np
-
 import theano.tensor as T
 
+from ..utils import int_types
+
 
 # 1D convolutions
 
@@ -60,7 +61,7 @@ def conv1d_mc0(input, filters, image_shape=None, filter_shape=None,
 
  if isinstance(border_mode, tuple):
  (border_mode,) = border_mode
- if isinstance(border_mode, int):
+ if isinstance(border_mode, int_types):
  border_mode = (0, border_mode)
 
  input_mc0 = input.dimshuffle(0, 1, 'x', 2)
@@ -94,7 +95,7 @@ def conv1d_mc1(input, filters, image_shape=None, filter_shape=None,
 
  if isinstance(border_mode, tuple):
  (border_mode,) = border_mode
- if isinstance(border_mode, int):
+ if isinstance(border_mode, int_types):
  border_mode = (border_mode, 0)
 
  input_mc1 = input.dimshuffle(0, 1, 2, 'x')

lasagne/theano_extensions/padding.py

@@ -4,6 +4,8 @@
 
 import theano.tensor as T
 
+from ..utils import int_types
+
 
 def pad(x, width, val=0, batch_ndim=1):
  """
@@ -33,7 +35,7 @@ def pad(x, width, val=0, batch_ndim=1):
  output_shape = list(input_shape)
  indices = [slice(None) for _ in output_shape]
 
- if isinstance(width, int):
+ if isinstance(width, int_types):
  widths = [width] * (input_ndim - batch_ndim)
  else:
  widths = width

lasagne/utils.py

@@ -1,9 +1,15 @@
-import numpy as np
+import numbers
 
+import numpy as np
 import theano
 import theano.tensor as T
 
 
+#: Tuple of ``int``-like types for ``isinstance`` checks.
+#: Specifically includes long integers and numpy integers.
+int_types = (numbers.Integral, np.integer)
+
+
 def floatX(arr):
  """Converts data to a numpy array of dtype ``theano.config.floatX``.
 
@@ -171,8 +177,8 @@ def as_tuple(x, N, t=None):
  x : value or iterable
  N : integer
  length of the desired tuple
- t : type, optional
- required type for all elements
+ t : type or tuple of type, optional
+ required type or types for all elements
 
  Returns
  -------
@@ -192,8 +198,14 @@ def as_tuple(x, N, t=None):
  X = (x,) * N
 
  if (t is not None) and not all(isinstance(v, t) for v in X):
+ if t == int_types:
+ expected_type = "int" # easier to understand
+ elif isinstance(t, tuple):
+ expected_type = " or ".join(tt.__name__ for tt in t)
+ else:
+ expected_type = t.__name__
  raise TypeError("expected a single value or an iterable "
- "of {0}, got {1} instead".format(t.__name__, x))
+ "of {0}, got {1} instead".format(expected_type, x))
 
  if len(X) != N:
  raise ValueError("expected a single value or an iterable "