本文整理汇总了Python中theano.tensor.switch方法的典型用法代码示例。如果您正苦于以下问题:Python tensor.switch方法的具体用法?Python tensor.switch怎么用?Python tensor.switch使用的例子?那么, 这里精选的方法代码示例或许可以为您提供帮助。您也可以进一步了解该方法所在类theano.tensor
的用法示例。
在下文中一共展示了tensor.switch方法的15个代码示例,这些例子默认根据受欢迎程度排序。您可以为喜欢或者感觉有用的代码点赞,您的评价将有助于系统推荐出更棒的Python代码示例。
示例1: reduce_log_sum
# 需要导入模块: from theano import tensor [as 别名]
# 或者: from theano.tensor import switch [as 别名]
def reduce_log_sum(tensor, axis=None, guaranteed_finite=False):
"""
Sum probabilities in the log domain, i.e return
log(e^vec[0] + e^vec[1] + ...)
= log(e^x e^(vec[0]-x) + e^x e^(vec[1]-x) + ...)
= log(e^x [e^(vec[0]-x) + e^(vec[1]-x) + ...])
= log(e^x) + log(e^(vec[0]-x) + e^(vec[1]-x) + ...)
= x + log(e^(vec[0]-x) + e^(vec[1]-x) + ...)
For numerical stability, we choose x = max(vec)
Note that if x is -inf, that means all values are -inf,
so the answer should be -inf. In this case, choose x = 0
"""
maxval = T.max(tensor, axis)
maxval_full = T.max(tensor, axis, keepdims=True)
if not guaranteed_finite:
maxval = T.switch(T.isfinite(maxval), maxval, T.zeros_like(maxval))
maxval_full = T.switch(T.isfinite(maxval_full), maxval_full, T.zeros_like(maxval_full))
reduced_sum = T.sum(T.exp(tensor - maxval_full), axis)
logsum = maxval + T.log(reduced_sum)
return logsum
示例2: time_distributed_nonzero_max_pooling
# 需要导入模块: from theano import tensor [as 别名]
# 或者: from theano.tensor import switch [as 别名]
def time_distributed_nonzero_max_pooling(x):
"""
Computes maximum along the first (time) dimension.
It ignores the mask m.
In:
x - input; a 3D tensor
mask_value - value to mask out, if None then no masking;
by default 0.0,
"""
import theano.tensor as T
mask_value=0.0
x = T.switch(T.eq(x, mask_value), -numpy.inf, x)
masked_max_x = x.max(axis=1)
# replace infinities with mask_value
masked_max_x = T.switch(T.eq(masked_max_x, -numpy.inf), 0, masked_max_x)
return masked_max_x
示例3: time_distributed_masked_max
# 需要导入模块: from theano import tensor [as 别名]
# 或者: from theano.tensor import switch [as 别名]
def time_distributed_masked_max(x, m):
"""
Computes max along the first (time) dimension.
In:
x - input; a 3D tensor
m - mask
m_value - value for masking
"""
# place infinities where mask is off
m_value = 0.0
tmp = K.switch(K.equal(m, 0.0), -numpy.inf, 0.0)
x_with_inf = x + K.expand_dims(tmp)
x_max = K.max(x_with_inf, axis=1)
r = K.switch(K.equal(x_max, -numpy.inf), m_value, x_max)
return r
## classes ##
# Transforms existing layers to masked layers
示例4: __init__
# 需要导入模块: from theano import tensor [as 别名]
# 或者: from theano.tensor import switch [as 别名]
def __init__(self, scale_grads=1, target_scale=.1,
discriminator_default_input_include_prob = 1.,
discriminator_input_include_probs=None,
discriminator_default_input_scale=1.,
discriminator_input_scales=None,
generator_default_input_include_prob = 1.,
generator_default_input_scale=1.,
inference_default_input_include_prob=None,
inference_input_include_probs=None,
inference_default_input_scale=1.,
inference_input_scales=None,
init_now_train_generator=True,
ever_train_discriminator=True,
ever_train_generator=True,
ever_train_inference=True,
no_drop_in_d_for_g=False,
alternate_g = False):
self.__dict__.update(locals())
del self.self
# These allow you to dynamically switch off training parts.
# If the corresponding ever_train_* is False, these have
# no effect.
self.now_train_generator = sharedX(init_now_train_generator)
self.now_train_discriminator = sharedX(numpy.array(1., dtype='float32'))
self.now_train_inference = sharedX(numpy.array(1., dtype='float32'))
示例5: shared_dropout_layer
# 需要导入模块: from theano import tensor [as 别名]
# 或者: from theano.tensor import switch [as 别名]
def shared_dropout_layer(shape, use_noise, trng, value, scaled=True):
#re-scale dropout at training time, so we don't need to at test time
if scaled:
proj = tensor.switch(
use_noise,
trng.binomial(shape, p=value, n=1,
dtype='float32')/value,
theano.shared(numpy.float32(1.)))
else:
proj = tensor.switch(
use_noise,
trng.binomial(shape, p=value, n=1,
dtype='float32'),
theano.shared(numpy.float32(value)))
return proj
# feedforward layer: affine transformation + point-wise nonlinearity
示例6: AdaMax
# 需要导入模块: from theano import tensor [as 别名]
# 或者: from theano.tensor import switch [as 别名]
def AdaMax(w, objective, alpha=.01, beta1=.1, beta2=.001):
print 'AdaMax', 'alpha:',alpha,'beta1:',beta1,'beta2:',beta2
g = T.grad(objective.sum(), w, disconnected_inputs='warn')
new = OrderedDict()
for i in range(len(w)):
#gi = T.switch(T.isnan(gi),T.zeros_like(gi),gi) #remove NaN's
mom1 = G.sharedf(w[i].get_value() * 0.)
_max = G.sharedf(w[i].get_value() * 0.)
new[mom1] = (1-beta1) * mom1 + beta1 * g[i]
new[_max] = T.maximum((1-beta2)*_max, abs(g[i]) + 1e-8)
new[w[i]] = w[i] + alpha * new[mom1] / new[_max]
return new
# AdaMax that averages over multiple minibatches
示例7: sgdmgc
# 需要导入模块: from theano import tensor [as 别名]
# 或者: from theano.tensor import switch [as 别名]
def sgdmgc(cost, params, lr=1.0, alpha=0.1, max_magnitude=5.0, infDecay=0.1):
"""SGD with momentum and gradient clipping"""
grads = T.grad(cost=cost, wrt=params)
updates = []
norm = norm_gs(params, grads)
sqrtnorm = T.sqrt(norm)
not_finite = T.or_(T.isnan(sqrtnorm), T.isinf(sqrtnorm))
adj_norm_gs = T.switch(T.ge(sqrtnorm, max_magnitude), max_magnitude / sqrtnorm, 1.)
for p, g in zip(params, grads):
v = shared(p.get_value() * 0.)
g = T.switch(not_finite, infDecay * p, g * adj_norm_gs)
v_new = v * (1.0 - alpha) - alpha * lr * g
updates.append((v, v_new))
updates.append((p, p + v_new ))
return updates, norm
示例8: switch
# 需要导入模块: from theano import tensor [as 别名]
# 或者: from theano.tensor import switch [as 别名]
def switch(condition, then_expression, else_expression):
"""Switches between two operations depending on a scalar value.
Note that both `then_expression` and `else_expression`
should be symbolic tensors of the *same shape*.
# Arguments
condition: scalar tensor (`int` or `bool`).
then_expression: either a tensor, or a callable that returns a tensor.
else_expression: either a tensor, or a callable that returns a tensor.
# Returns
The selected tensor.
"""
if callable(then_expression):
then_expression = then_expression()
if callable(else_expression):
else_expression = else_expression()
cond_ndim = ndim(condition)
expr_ndim = ndim(then_expression)
if cond_ndim < expr_ndim:
ndim_diff = expr_ndim - cond_ndim
for _ in range(ndim_diff):
condition = expand_dims(condition)
return T.switch(condition, then_expression, else_expression)
示例9: bound
# 需要导入模块: from theano import tensor [as 别名]
# 或者: from theano.tensor import switch [as 别名]
def bound(expression, out, *predicates):
"""Bound a theano expression.
Parameters
----------
* `expression` [theano expression]:
The expression to bound.
* `out` [theano expression]:
The out-of-bounds value.
* `*predicates` [list of theano expressions]:
The list of predicates defining the boundaries of `expression`.
Returns
-------
* `value` [theano expression]:
The bounded expression.
"""
guard = 1
for p in predicates:
guard *= p
return T.switch(guard, expression, out)
示例10: reScale
# 需要导入模块: from theano import tensor [as 别名]
# 或者: from theano.tensor import switch [as 别名]
def reScale(self,energy,threshold,replaceValue=1e-7):
#{{{
assert energy.ndim==1;
maxValue=energy.max();
def checkThreshold(value,threshold,replaceValue):
return T.switch(T.lt(value,threshold),replaceValue,value);
result,update=theano.scan(fn=checkThreshold,
outputs_info=None,
sequences=[energy],
non_sequences=[threshold,replaceValue]);
return T.switch(T.lt(maxValue,threshold),energy,result);
#}}}
示例11: switch
# 需要导入模块: from theano import tensor [as 别名]
# 或者: from theano.tensor import switch [as 别名]
def switch(condition, then_expression, else_expression):
'''condition: scalar tensor.
'''
return T.switch(condition, then_expression, else_expression)
示例12: local_gpua_row_switch
# 需要导入模块: from theano import tensor [as 别名]
# 或者: from theano.tensor import switch [as 别名]
def local_gpua_row_switch(node):
"""
Detects eligible Switch instances and replaces them with a GPU
row switch.
"""
if (node.op.__class__ == T.Elemwise
and node.op.scalar_op.__class__ != theano.scalar.Switch):
return False
cond, ift, iff = node.inputs
out, = node.outputs
# Only applies to Switch instances where a vector mask broadcasts over
# matrices.
bcast = cond.broadcastable
if not bcast or not (not bcast[0] and all(bcast[1:])
and ift.ndim in [2, 3]):
return False
if not (ift.dtype == iff.dtype == "float32"):
return False
if cond.owner and isinstance(cond.owner.op, HostFromGpu):
gpu_cond, = cond.owner.inputs
else:
gpu_cond = as_cuda_ndarray_variable(
T.cast(cond.flatten(), "float32"))
if ift.owner and isinstance(ift.owner.op, HostFromGpu):
gpu_ift, = ift.owner.inputs
else:
gpu_ift = as_cuda_ndarray_variable(ift)
if iff.owner and isinstance(iff.owner.op, HostFromGpu):
gpu_iff, = iff.owner.inputs
else:
gpu_iff = as_cuda_ndarray_variable(iff)
gpu_op = GpuRowSwitch()
return [HostFromGpu()(gpu_op(cond, gpu_ift, gpu_iff))]
示例13: fprop
# 需要导入模块: from theano import tensor [as 别名]
# 或者: from theano.tensor import switch [as 别名]
def fprop(self, state_below):
p = state_below
p = T.switch(p > 0., p, self.left_slope * p)
return p
示例14: switch
# 需要导入模块: from theano import tensor [as 别名]
# 或者: from theano.tensor import switch [as 别名]
def switch(condition, then_expression, else_expression):
'''condition: scalar tensor.
'''
return T.switch(condition, then_expression, else_expression)
# NN OPERATIONS
示例15: clip_norm
# 需要导入模块: from theano import tensor [as 别名]
# 或者: from theano.tensor import switch [as 别名]
def clip_norm(g, c, n):
if c > 0:
g = T.switch(T.ge(n, c), g * c / n, g)
return g