EderSantana

"""
Multilayer Perceptron
"""
__authors__ = "Ian Goodfellow"
__copyright__ = "Copyright 2012-2013, Universite de Montreal"
__credits__ = ["Ian Goodfellow", "David Warde-Farley"]
__license__ = "3-clause BSD"
__maintainer__ = "Ian Goodfellow"

import math

EderSantana

!obj:pylearn2.train.Train {
    dataset: &train !obj:pylearn2.datasets.mnist.MNIST {
        which_set: 'train',
        one_hot: 1,
        start: 0,
        stop: 50000
    },
    model: !obj:pylearn2.models.mlp.MLP {
        batch_size: 128,
        layers: [

EderSantana

!obj:pylearn2.train.Train {
    dataset: &train !obj:pylearn2.datasets.mnist.MNIST {
        which_set: 'train',
        one_hot: 1,
        axes: ['c', 0, 1, 'b'],
        start: 0,
        stop: 50000
    },
    model: !obj:pylearn2.models.mlp.MLP {
        batch_size: 128,

EderSantana

EderSantana

EderSantana

EderSantana

class ConditionedRecurrent(BaseRecurrent):

    def __init__(self, wrapped, **kwargs):
        super(ConditionedRecurrent, self).__init__(**kwargs)
        self.wrapped = wrapped
        self.children=[wrapped]

    def get_dim(self, name):
        if name == 'context':
            return self.wrapped.get_dim('inputs')

EderSantana

# You are supposed to run this after the `Batch Normalization Tutorial`
import theano
import numpy as np

outputs = VariableFilter(
             bricks=mlp.linear_transformations, roles=[OUTPUT])(cg_bn.variables)

f = []
for o, g, b in zip(outputs, gammas, betas):
    f.append(

EderSantana

# We are training this DRAW network
# using a model similar to the one 
# described in the paper http://arxiv.org/pdf/1502.04623.pdf
#
# Dataset: Binary-MNIST from mila-udem/fuel
from keras.initializations import normal
from seya.layers.draw import DRAW

def myinit(shape):
    return normal(shape, scale=.01)

EderSantana

# NOTE: I'm not sure if this is right

from keras.layers.recurrent import LSTM

class LSTMpeephole(LSTM):
    def __init__(self, **kwargs):
        super(LSTMpeephole, self).__init__(**kwargs)

    def build(self):
        super(LSTMpeephole, self).build()