Conversion of the cifar10.yaml from pylearn2 - https://github.com/lisa-lab/pylearn2/blob/master/pylearn2/scripts/papers/maxout/cifar10.yaml

cifar10.py

from pylearn2.utils import serial

from pylearn2.datasets.zca_dataset import ZCA_Dataset
dataset = ZCA_Dataset(
    preprocessed_dataset=serial.load("${PYLEARN2_DATA_PATH}/cifar10/pylearn2_gcn_whitened/train.pkl"),
    preprocessor=serial.load("${PYLEARN2_DATA_PATH}/cifar10/pylearn2_gcn_whitened/preprocessor.pkl"),
    axes=['c', 0, 1, 'b']
)

from pylearn2.models.maxout import MaxoutConvC01B
h0 = MaxoutConvC01B(
    layer_name='h0',
    pad=4,
    tied_b=1,
    W_lr_scale=.05,
    b_lr_scale=.05,
    num_channels=96,
    num_pieces=2,
    kernel_shape=[8, 8],
    pool_shape=[4, 4],
    pool_stride=[2, 2],
    irange=.005,
    max_kernel_norm=.9,
    partial_sum=33
)

h1 = MaxoutConvC01B(
    layer_name='h1',
    pad=3,
    tied_b=1,
    W_lr_scale=.05,
    b_lr_scale=.05,
    num_channels=192,
    num_pieces=2,
    kernel_shape=[8, 8],
    pool_shape=[4, 4],
    pool_stride=[2, 2],
    irange=.005,
    max_kernel_norm=1.9365,
    partial_sum=15
)

h2 = MaxoutConvC01B(
    pad=3,
    layer_name='h2',
    tied_b=1,
    W_lr_scale=.05,
    b_lr_scale=.05,
    num_channels=192,
    num_pieces=2,
    kernel_shape=[5, 5],
    pool_shape=[2, 2],
    pool_stride=[2, 2],
    irange=.005,
    max_kernel_norm=1.9365
)

from pylearn2.models.maxout import Maxout
h3 = Maxout(
    layer_name='h3',
    irange=.005,
    num_units=500,
    num_pieces=5,
    max_col_norm=1.9
)

from pylearn2.models.mlp import Softmax
y = Softmax(
    max_col_norm=1.9365,
    layer_name='y',
    n_classes=10,
    irange=.005
)

window_shape = [32, 32]

from pylearn2.space import Conv2DSpace
input_space = Conv2DSpace(
    shape=window_shape,
    num_channels=3,
    axes=['c', 0, 1, 'b'],
)

from pylearn2.models.mlp import MLP
model = MLP(layers=[h0, h1, h2, h3, y],
            input_space=input_space)

valid_dataset = ZCA_Dataset(
    preprocessed_dataset=serial.load("${PYLEARN2_DATA_PATH}/cifar10/pylearn2_gcn_whitened/test.pkl"),
    preprocessor=serial.load("${PYLEARN2_DATA_PATH}/cifar10/pylearn2_gcn_whitened/preprocessor.pkl"),
    axes=['c', 0, 1, 'b']
)

from pylearn2.training_algorithms.sgd import SGD
from pylearn2.training_algorithms.learning_rule import Momentum
from pylearn2.termination_criteria import EpochCounter
from pylearn2.costs.mlp.dropout import Dropout
algorithm = SGD(
    batch_size=128,
    learning_rate=.17,
    learning_rule=Momentum(init_momentum=.5),
    monitoring_dataset={'test': valid_dataset},
    cost=Dropout(
        input_include_probs={'h0': .8},
        input_scales={'h0': 1.}
    ),
    termination_criterion=EpochCounter(max_epochs=474)
)

from pylearn2.train import Train
from pylearn2.training_algorithms.learning_rule import MomentumAdjustor
from pylearn2.training_algorithms.sgd import LinearDecayOverEpoch
from pylearn2.train_extensions.window_flip import WindowAndFlip
train_obj = Train(
    dataset=dataset,
    model=model,
    algorithm=algorithm,
    extensions=[
        MomentumAdjustor(
            start=1,
            saturate=250,
            final_momentum=.65
        ),
        LinearDecayOverEpoch(
            start=1,
            saturate=500,
            decay_factor=.01
        ),
        WindowAndFlip(
            pad_randomized=4,
            window_shape=window_shape,
            randomize=[dataset],
            center=[valid_dataset]
        )
    ],
    save_path="${PYLEARN2_TRAIN_FILE_FULL_STEM}.pkl",
    save_freq=1
)

train_obj.main_loop()