def sum(x, axis=None):
if isinstance(x, list):
x = tf.pack(x)
if axis is None:
result = tf.reduce_sum(x)
result_shape = []
else:
if axis < 0:
axis = x.ndim + axis
result = tf.reduce_sum(x, axis)
result_shape = get_raw_dimensions(x)
result_shape = list(result_shape[:axis]) + list(result_shape[axis + 1:])
if get_raw_dimensions(result).ndims is None:
result.set_shape(result_shape)
return result
def sum(x, axis=None):
if isinstance(x, list):
x = tf.pack(x)
if axis is None:
result = tf.reduce_sum(x)
result_shape = []
else:
if axis < 0:
axis = x.ndim + axis
result = tf.reduce_sum(x, axis)
result_shape = get_raw_dimensions(x)
result_shape = list(result_shape[:axis]) + list(
result_shape[axis + 1:])
if get_raw_dimensions(result).ndims is None:
result.set_shape(result_shape)
return result
def _tf_sub(self, other):
self_dim = list(get_raw_dimensions(self))
other_dim = list(get_raw_dimensions(other))
if isinstance(self, tf.Variable):
self = self._AsTensor()
if isinstance(other, tf.Variable):
other = other._AsTensor()
result = _old_sub(self, other)
result_dim = get_raw_dimensions(result)
if result_dim.ndims is None:
# we could infer the shape in this case
if len(self_dim) > len(other_dim):
result.set_shape(self_dim)
else:
result.set_shape(other_dim)
return result
def _tf_sub(self, other):
self_dim = list(get_raw_dimensions(self))
other_dim = list(get_raw_dimensions(other))
if isinstance(self, tf.Variable):
self = self._AsTensor()
if isinstance(other, tf.Variable):
other = other._AsTensor()
result = _old_sub(self, other)
result_dim = get_raw_dimensions(result)
if result_dim.ndims is None:
# we could infer the shape in this case
if len(self_dim) > len(other_dim):
result.set_shape(self_dim)
else:
result.set_shape(other_dim)
return result
def _tf_rmul(self, other):
self_dim = list(get_raw_dimensions(self))
other_dim = list(get_raw_dimensions(other))
if isinstance(self, tf.Variable):
self = self._AsTensor()
if isinstance(other, tf.Variable):
other = other._AsTensor()
if not self.dtype.is_floating and (isinstance(other, float) or other.dtype.is_floating):
self = tf.cast(self, tf.float32)
result = _old_rmul(self, other)
result_dim = get_raw_dimensions(result)
if result_dim.ndims is None:
# we could infer the shape in this case
if len(self_dim) > len(other_dim):
result.set_shape(self_dim)
else:
result.set_shape(other_dim)
return result
def _tf_rmul(self, other):
self_dim = list(get_raw_dimensions(self))
other_dim = list(get_raw_dimensions(other))
if isinstance(self, tf.Variable):
self = self._AsTensor()
if isinstance(other, tf.Variable):
other = other._AsTensor()
if not self.dtype.is_floating and (isinstance(other, float)
or other.dtype.is_floating):
self = tf.cast(self, tf.float32)
result = _old_rmul(self, other)
result_dim = get_raw_dimensions(result)
if result_dim.ndims is None:
# we could infer the shape in this case
if len(self_dim) > len(other_dim):
result.set_shape(self_dim)
else:
result.set_shape(other_dim)
return result
def exp(x):
result = tf.exp(x)
if get_raw_dimensions(result).ndims is None:
result.set_shape(get_raw_dimensions(x))
return result
def exp(x):
result = tf.exp(x)
if get_raw_dimensions(result).ndims is None:
result.set_shape(get_raw_dimensions(x))
return result