def __init__(self): self.numpy_rng = numpy.random.RandomState(89677) self.theano_rng = RandomStreams(self.numpy_rng.randint(2**30)) self.x1 = T.tensor3('x1') self.x2 = T.tensor3('x2') self.x1_bis = T.matrix('x1_bis') self.x2_bis = T.matrix('x2_bis') self.y = T.vector('y') #0 or 1 self.z = T.vector('z') #-1 or 1 self.model1 = NN_layers.doubleDotMeanLayer(self.numpy_rng, self.theano_rng, self.x1, self.x1_bis, INPUT_SHAPE, n_out=OUTPUT_SHAPE[1], W=None) self.model2 = NN_layers.doubleDotMeanLayer(self.numpy_rng, self.theano_rng, self.x2, self.x2_bis, INPUT_SHAPE, n_out=OUTPUT_SHAPE[1], W=self.model1.W) self.params = [] self.params.extend(self.model1.params)
def __init__(self): self.numpy_rng = numpy.random.RandomState(89677) self.theano_rng = RandomStreams(self.numpy_rng.randint(2**30)) self.x1 = T.tensor3('x1') self.x2 = T.tensor3('x2') self.l1 = T.vector('l1') self.l2 = T.vector('l2') self.y = T.vector('y') #0 or 1 self.z = T.vector('z') #-1 or 1 self.model1 = NN_layers.exponentialWeightsDotMeanLayer( self.numpy_rng, self.theano_rng, self.x1, INPUT_SHAPE, self.l1, max_length=35, n_out=OUTPUT_SHAPE[1], batch_size=BATCH_SIZE, W=None) self.model2 = NN_layers.exponentialWeightsDotMeanLayer( self.numpy_rng, self.theano_rng, self.x2, INPUT_SHAPE, self.l2, max_length=35, n_out=OUTPUT_SHAPE[1], batch_size=BATCH_SIZE, W=self.model1.W) self.params = [] self.params.extend(self.model1.params)
def __init__(self): self.numpy_rng = numpy.random.RandomState(89677) self.theano_rng = RandomStreams(self.numpy_rng.randint(2 ** 30)) self.x1 = T.tensor3('x1') self.x2 = T.tensor3('x2') self.y = T.vector('y') #0 or 1 self.z = T.vector('z') #-1 or 1 self.model1 = NN_layers.MLPLayer(self.numpy_rng, self.theano_rng, self.x1, INPUT_SHAPE, n_out=OUTPUT_SHAPE[1], batch_size=BATCH_SIZE, W=None, b=None) self.model2 = NN_layers.MLPLayer(self.numpy_rng, self.theano_rng, self.x2, INPUT_SHAPE, n_out=OUTPUT_SHAPE[1], batch_size=BATCH_SIZE, W=self.model1.W, b=self.model1.b) self.params = [] self.params.extend(self.model1.params)
def __init__(self): self.numpy_rng = numpy.random.RandomState(89677) self.theano_rng = RandomStreams(self.numpy_rng.randint(2 ** 30)) self.x1 = T.tensor3('x1') self.x2 = T.tensor3('x2') self.y = T.vector('y') #0 or 1 self.z = T.vector('z') #-1 or 1 self.model1 = NN_layers.dotMeanLayer(self.numpy_rng, self.theano_rng, self.x1, INPUT_SHAPE, n_out=OUTPUT_SHAPE[1], W=None) self.model2 = NN_layers.dotMeanLayer(self.numpy_rng, self.theano_rng, self.x2, INPUT_SHAPE, n_out=OUTPUT_SHAPE[1], W=self.model1.W) self.params = [] self.params.extend(self.model1.params)
def __init__(self): self.numpy_rng = numpy.random.RandomState(89677) self.theano_rng = RandomStreams(self.numpy_rng.randint(2 ** 30)) self.x1 = T.tensor3('x1') self.x2 = T.tensor3('x2') self.l1 = T.vector('l1') self.l2 = T.vector('l2') self.y = T.vector('y') #0 or 1 self.z = T.vector('z') #-1 or 1 self.model1 = NN_layers.quarticWeightsDotMeanLayer(self.numpy_rng, self.theano_rng, self.x1, INPUT_SHAPE, self.l1, max_length=35, n_out=OUTPUT_SHAPE[1], batch_size=BATCH_SIZE, W=None) self.model2 = NN_layers.quarticWeightsDotMeanLayer(self.numpy_rng, self.theano_rng, self.x2, INPUT_SHAPE, self.l2, max_length=35, n_out=OUTPUT_SHAPE[1], batch_size=BATCH_SIZE, W=self.model1.W) self.params = [] self.params.extend(self.model1.params)
def __init__(self): self.numpy_rng = numpy.random.RandomState(89677) self.theano_rng = RandomStreams(self.numpy_rng.randint(2 ** 30)) self.x1 = T.tensor3("x1") self.x2 = T.tensor3("x2") self.indices1 = T.matrix("i1") self.indices2 = T.matrix("i2") self.l1 = T.vector("l1") self.l2 = T.vector("l2") self.y = T.vector("y") # 0 or 1 self.z = T.vector("z") # -1 or 1 self.model1 = NN_layers.exponentialTweetDotMeanLayer( self.numpy_rng, self.theano_rng, self.x1, INPUT_SHAPE, self.indices1, self.l1, max_length=35, n_out=OUTPUT_SHAPE[1], batch_size=BATCH_SIZE, W=None, ) self.model2 = NN_layers.exponentialTweetDotMeanLayer( self.numpy_rng, self.theano_rng, self.x2, INPUT_SHAPE, self.indices2, self.l2, max_length=35, n_out=OUTPUT_SHAPE[1], batch_size=BATCH_SIZE, W=self.model1.W, ) self.params = [] self.params.extend(self.model1.params)
def model(x): x = tf.reshape(x, [-1, 784]) return geebs_tanh(layer.fcl(x, [784, 100, 100, 10], "FCL1"))