# normal/centered RBM --> 0.0/0.01
update_offsets = 0.01
# Flipped/Normal MNIST --> True/False
flipped = False
# Set random seed (optional)
numx.random.seed(42)
# Input and hidden dimensionality
v1 = v2 = 28
h1 = h2 = 4
# Load data (download is not existing)
train_data, _, valid_data, _, test_data, _ = io.load_mnist(
"mnist.pkl.gz", True)
train_data = numx.vstack((train_data, valid_data))
# Flip the dataset if chosen
if flipped:
train_data = 1 - train_data
test_data = 1 - test_data
print("Flipped MNIST")
else:
print("Normal MNIST")
# Training parameters
batch_size = 100
epochs = 50
# Create centered or normal model
import numpy as numx
import pydeep.fnn.model as MODEL
import pydeep.fnn.layer as LAYER
import pydeep.fnn.trainer as TRAINER
import pydeep.base.activationfunction as ACT
import pydeep.base.costfunction as COST
import pydeep.base.corruptor as CORR
import pydeep.misc.io as IO
import pydeep.base.numpyextension as npExt
# Set random seed (optional)
numx.random.seed(42)
# Load data and whiten it
train_data, train_label, valid_data, valid_label, test_data, test_label = IO.load_mnist(
"mnist.pkl.gz", False)
train_data = numx.vstack((train_data, valid_data))
train_label = numx.hstack((train_label, valid_label)).T
train_label = npExt.get_binary_label(train_label)
test_label = npExt.get_binary_label(test_label)
# Create model
l1 = LAYER.FullConnLayer(input_dim=train_data.shape[1],
output_dim=1000,
activation_function=ACT.ExponentialLinear(),
initial_weights='AUTO',
initial_bias=0.0,
initial_offset=numx.mean(train_data, axis=0).reshape(
1, train_data.shape[1]),
connections=None,
dtype=numx.float64)
import pydeep.misc.visualization as VIS
import pydeep.misc.io as IO
import pydeep.base.numpyextension as numxExt
from pydeep.dbm.unit_layer import *
from pydeep.dbm.weight_layer import *
from pydeep.dbm.model import *
# Set the same seed value for all algorithms
numx.random.seed(42)
# Load Data
train_data = IO.load_mnist("mnist.pkl.gz", True)[0]
# Set dimensions Layer 1-3
v11 = v12 = 28
v21 = v22 = 10
v31 = v32 = 10
N = v11 * v12
M = v21 * v22
O = v31 * v32
# Create weight layers, which connect the unit layers
wl1 = Weight_layer(input_dim=N,
output_dim=M,
initial_weights=0.01,
dtype=numx.float64)
wl2 = Weight_layer(input_dim=M,
output_dim=O,
initial_weights=0.01,
dtype=numx.float64)
# Import cost functions, activation function, Autencoder and trainer module
import pydeep.base.activationfunction as act
import pydeep.base.costfunction as cost
import pydeep.ae.model as aeModel
import pydeep.ae.trainer as aeTrainer
# Set random seed (optional)
numx.random.seed(42)
# Input and hidden dimensionality
v1 = v2 = 28
h1 = 10
h2 = 10
# Load data , get it from 'deeplearning.net/data/mnist/mnist.pkl.gz'
train_data, _, _, _, test_data, _ = io.load_mnist("../../data/mnist.pkl.gz",
False)
# Set hyperparameters batchsize and number of epochs
batch_size = 10
max_epochs = 10
# Create model with sigmoid hidden units, linear output units, and squared error loss.
ae = aeModel.AutoEncoder(v1 * v2,
h1 * h2,
data=train_data,
visible_activation_function=act.Sigmoid(),
hidden_activation_function=act.Sigmoid(),
cost_function=cost.CrossEntropyError(),
initial_weights='AUTO',
initial_visible_bias='AUTO',
initial_hidden_bias='AUTO',