def __init__(self,
n_visible,
n_hidden,
train_k=1,
sample_k=1000,
theano_rng=None,
persistent=None,
**kwargs):
self.n_visible = n_visible
self.n_hidden = n_hidden
if theano_rng is None:
theano_rng = RandomStreams(np.random.randint(2**30))
assert kwargs.has_key('weight')
assert kwargs.has_key('hbias')
assert kwargs.has_key('vbias')
assert not kwargs['weight'].has_key('shape')
assert not kwargs['vbias'].has_key('shape')
assert not kwargs['hbias'].has_key('shape')
self.W = init_shared(shape=(n_visible, n_hidden), **kwargs['weight'])
self.hbias = init_shared(shape=(n_hidden, ), **kwargs['hbias'])
self.vbias = init_shared(shape=(n_visible, ), **kwargs['vbias'])
self.theano_rng = theano_rng
self.persistent = persistent
self.train_k = train_k
self.sample_k = sample_k
self.input = T.matrix()
def __init__(self, subSampleShape, **kwargs):
layer.__init__(self)
assert len(subSampleShape) == 2
self.subSampleShape = subSampleShape
assert kwargs.has_key('coef')
assert kwargs.has_key('bias')
self.coef = init_shared(**kwargs['coef'])
self.bias = init_shared(**kwargs['bias'])
def __init__(self, subSampleShape, **kwargs):
layer.__init__(self)
assert len(subSampleShape) == 2
self.subSampleShape = subSampleShape
assert kwargs.has_key('coef')
assert kwargs.has_key('bias')
self.coef = init_shared(**kwargs['coef'])
self.bias = init_shared(**kwargs['bias'])
def __init__(self,
n_visible, n_hidden, train_k=1, sample_k=1000,
theano_rng=None, persistent=None,
**kwargs):
self.n_visible = n_visible
self.n_hidden = n_hidden
if theano_rng is None:
theano_rng = RandomStreams(np.random.randint(2**30))
assert kwargs.has_key('weight')
assert kwargs.has_key('hbias')
assert kwargs.has_key('vbias')
assert not kwargs['weight'].has_key('shape')
assert not kwargs['vbias'].has_key('shape')
assert not kwargs['hbias'].has_key('shape')
self.W = init_shared(shape=(n_visible, n_hidden), **kwargs['weight'] )
self.hbias = init_shared(shape=(n_hidden,), **kwargs['hbias'] )
self.vbias = init_shared(shape=(n_visible,), **kwargs['vbias'] )
self.theano_rng = theano_rng
self.persistent = persistent
self.train_k = train_k
self.sample_k = sample_k
self.input = T.matrix()
def init_bias(self):
self.b = init_shared(shape=(self.numOfOutput, ), **self.biasKwargs)
def init_weights(self):
self.W = init_shared(shape=self.get_weightShape(), **self.weightKwargs)
def init_kernels(self, numOfConnections):
self.kernels = init_shared(shape=numOfConnections, **self.kernelKwargs)
def init_kernels(self, numOfConnections):
self.kernels = init_shared(shape=numOfConnections, **self.kernelKwargs)
def init_filters(self):
self.filters = []
for i in xrange(self.get_numOfFilters()):
self.filters.append(
init_shared(shape=self.get_filterShape(), **self.filterKwargs))
def init_bias(self):
self.bias = init_shared(**self.biasKwargs)
def init_filters(self, n_pre_fm):
self.filterShape = (self.n_fm, n_pre_fm) + self.filterDs
self.filters = init_shared(shape=self.get_filterShape(),
**self.filterKwargs)
def init_bias(self):
'''
:var n_fm means the number of this layer's feature maps
'''
self.bias = init_shared(shape=(self.n_fm, ), **self.biasKwargs)
def init_filters(self, n_pre_fm):
self.filterShape = (self.n_fm, n_pre_fm) + self.filterDs
self.filters = init_shared(shape=self.get_filterShape(), **self.filterKwargs)
def init_bias(self):
'''
:var n_fm means the number of this layer's feature maps
'''
self.bias = init_shared(shape=(self.n_fm,), **self.biasKwargs)
def init_filters(self):
self.filters = []
for i in xrange(self.get_numOfFilters()):
self.filters.append( init_shared(shape=self.get_filterShape(), **self.filterKwargs) )
def init_bias(self):
self.bias = init_shared(**self.biasKwargs)
