Max-Planck-Institute for Medical Research, Heidelberg, Germany
Authors: Marius Killinger, Gregor Urban
"""
import matplotlib.pyplot as plt
from elektronn.training.traindata import MNISTData
from elektronn.net.convnet import MixedConvNN
from elektronn.net.introspection import embedMatricesInGray
# Load Data #
data = MNISTData(path=None, convert2image=False, shift_augment=False)
# Create Autoencoder #
batch_size = 100
cnn = MixedConvNN((28**2), input_depth=None)
cnn.addPerceptronLayer(n_outputs=300, activation_func="tanh")
cnn.addPerceptronLayer(n_outputs=200, activation_func="tanh")
cnn.addPerceptronLayer(n_outputs=50, activation_func="tanh")
cnn.addTiedAutoencoderChain(n_layers=None,
activation_func="tanh",
input_noise=0.3)
cnn.compileOutputFunctions(
target="regression") #compiles the cnn.get_error function as well
cnn.setOptimizerParams(SGD={'LR': 5e-1, 'momentum': 0.9}, weight_decay=0)
print "training..."
for i in range(40000):
d, l = data.getbatch(batch_size)
loss, loss_instance, time_per_step = cnn.trainingStep(d, d, mode="SGD")
# -*- coding: utf-8 -*-
"""
ELEKTRONN - Neural Network Toolkit
Copyright (c) 2015 Gregor Urban, Marius Killinger
"""
from elektronn.training.traindata import MNISTData
from elektronn.net.convnet import MixedConvNN
# Load Data #
data = MNISTData(path=None, convert2image=True, shift_augment=False)
# Create CNN #
batch_size = 100
cnn = MixedConvNN((28, 28), input_depth=1, enable_dropout=True)
cnn.addConvLayer(10,
5,
pool_shape=2,
activation_func="tanh",
force_no_dropout=True) # (nof, filtersize)
cnn.addConvLayer(8,
5,
pool_shape=2,
activation_func="tanh",
force_no_dropout=True)
cnn.addPerceptronLayer(200, activation_func="tanh")
cnn.addPerceptronLayer(150, activation_func="tanh")
cnn.addPerceptronLayer(
10, activation_func="tanh", force_no_dropout=True
) # need 10 outputs as there are 10 classes in the data set
cnn.compileOutputFunctions()
"""
ELEKTRONN - Neural Network Toolkit
Copyright (c) 2015 Gregor Urban, Marius Killinger
"""
from elektronn.training.traindata import MNISTData
from elektronn.net.convnet import MixedConvNN
# Load Data #
data = MNISTData(path=None,
convert2image=True,
shift_augment=False)
# Create CNN #
batch_size = 100
cnn = MixedConvNN((28, 28), input_depth=1)
cnn.addConvLayer(10,
5,
pool_shape=2,
activation_func="abs") # (nof, filtersize)
cnn.addConvLayer(8, 5, pool_shape=2, activation_func="abs")
cnn.addPerceptronLayer(100, activation_func="abs")
cnn.addPerceptronLayer(80, activation_func="abs")
cnn.addPerceptronLayer(
10,
activation_func="abs"
) # need 10 outputs as there are 10 classes in the data set
cnn.compileOutputFunctions()
cnn.setOptimizerParams(SGD={'LR': 1e-2, 'momentum': 0.9}, weight_decay=0)
print "training..."
"""
import matplotlib.pyplot as plt
from elektronn.training.traindata import MNISTData
from elektronn.net.convnet import MixedConvNN
from elektronn.net.introspection import embedMatricesInGray
# Load Data #
data = MNISTData(path=None,
convert2image=False,
shift_augment=False)
# Create Autoencoder #
batch_size = 100
cnn = MixedConvNN((28**2), input_depth=None)
cnn.addPerceptronLayer(n_outputs=300, activation_func="tanh")
cnn.addPerceptronLayer(n_outputs=200, activation_func="tanh")
cnn.addPerceptronLayer(n_outputs=50, activation_func="tanh")
cnn.addTiedAutoencoderChain(n_layers=None,
activation_func="tanh",
input_noise=0.3)
cnn.compileOutputFunctions(target="regression") #compiles the cnn.get_error function as well
cnn.setOptimizerParams(SGD={'LR': 5e-1, 'momentum': 0.9}, weight_decay=0)
print "training..."
for i in range(40000):
d, l = data.getbatch(batch_size)
loss, loss_instance, time_per_step = cnn.trainingStep(d, d, mode="SGD")
if i % 100 == 0:
