from layer.core import *
from algorithm.SGD import Mini_Batch
from data.process import loadData
from layer.model import Model
if __name__ == '__main__':
dataSet=loadData()
cifar=Model(batch_size=100,lr=0.0001,dataSet=dataSet,weight_decay=0.004)
neure=[32,32,64,64]
batch_size=100
cifar.add(DataLayer(batch_size,(32,32,3)))
cifar.add(ConvolutionLayer((batch_size,3,32,32),(neure[0],3,3,3),'relu','Gaussian',0.0001))
cifar.add(PoolingLayer())
cifar.add(ConvolutionLayer((batch_size,neure[0],15,15),(neure[1],neure[0],4,4),'relu','Gaussian',0.01))
cifar.add(PoolingLayer())
cifar.add(ConvolutionLayer((batch_size,neure[1],6,6),(neure[2],neure[1],5,5),'relu','Gaussian',0.01))
cifar.add(PoolingLayer())
cifar.add(FullyConnectedLayer(neure[2]*1*1,neure[3],'relu','Gaussian',0.1))
cifar.add(DropoutLayer(0.5))
cifar.add(SoftmaxLayer(neure[3],5,'Gaussian',0.1))
cifar.build_train_fn()
cifar.build_vaild_fn()
algorithm=Mini_Batch(model=cifar,n_epochs=100,load_param='cnn_params.pkl',save_param='cnn_params.pkl')
algorithm.run()
from layer.core import *
from algorithm.SGD import Mini_Batch
from data.process import loadData, loadScaleData
from layer.model import Model
if __name__ == '__main__':
dataSet=loadScaleData('data.pkl')
cifar=Model(batch_size=100,lr=0.01,dataSet=dataSet,weight_decay=0.0)
neure=[1000,1000,1000]
batch_size=100
cifar.add(DataLayer(batch_size,32*32*3))
cifar.add(FullyConnectedLayer(32*32*3,neure[0],'relu','Gaussian',0.1))
cifar.add(DropoutLayer(0.2))
cifar.add(FullyConnectedLayer(neure[0],neure[1],'relu','Gaussian',0.1))
cifar.add(DropoutLayer(0.2))
cifar.add(FullyConnectedLayer(neure[1],neure[2],'relu','Gaussian',0.1))
cifar.add(DropoutLayer(0.2))
cifar.add(SoftmaxLayer(neure[2],10))
cifar.pretrain()
cifar.build_train_fn()
cifar.build_vaild_fn()
algorithm=Mini_Batch(model=cifar,n_epochs=100,load_param='mlp_params.pkl',save_param='mlp_params.pkl')
algorithm.run()
from layer.core import *
from algorithm.SGD import Mini_Batch
from data.process import loadTrainData
from layer.model import Model
if __name__ == '__main__':
# load ROI+ROTATION dataset
#dataSet=loadTrainData("dataset/ROT/data.pkl","dataset/ROT/mean.pkl",scale=128.0);
# load ROI dataset
dataSet=loadTrainData("dataset/ROI/data.pkl","dataset/ROI/mean.pkl",scale=128.0);
# load normal dataset
#dataSet=loadTrainData("dataset/NORMAL/data.pkl","dataset/NORMAL/mean.pkl",scale=128.0);
cifar=Model(batch_size=100,lr=0.001,dataSet=dataSet,weight_decay=0.0)
#neure=[2000]
neure=[1000,1000,1000]
#neure=[2000,2000,2000];
batch_size=100
cifar.add(DataLayer(batch_size,32*32))
cifar.add(FullyConnectedLayer(32*32,neure[0],'relu','Gaussian',0.1))
cifar.add(DropoutLayer(0.2))
cifar.add(FullyConnectedLayer(neure[0],neure[1],'relu','Gaussian',0.1))
cifar.add(DropoutLayer(0.2))
cifar.add(FullyConnectedLayer(neure[1],neure[2],'relu','Gaussian',0.1))
cifar.add(DropoutLayer(0.2))
cifar.add(SoftmaxLayer(neure[2],5))
cifar.build_train_fn()
cifar.build_vaild_fn()
algorithm=Mini_Batch(model=cifar,n_epochs=200,load_param='mlp_params.pkl',save_param='mlp_params.pkl')
algorithm.run()
def test_knn_stat():
print "\nNow tesing with KNN-ROI.....\n"
global test_hit, test_count
for idx in xrange(5):
test_knn(("dataset/ROI_TEST/%d.bin") % idx, "dataset/ROI/mean.pkl",
idx)
print "total error rate: %f%%\n" % (float(test_count - test_hit) /
test_count * 100)
# clear
test_hit = test_count = 0
if __name__ == '__main__':
Test_Single = Model(batch_size=1, lr=0.01, dataSet=None)
meta_num = 100
neure = [meta_num, meta_num, meta_num, meta_num]
batch_size = 1
x = T.matrix('x')
index = T.lscalar()
Test_Single.add(DataLayer(batch_size, (32, 32, 1)))
Test_Single.add(
ConvolutionLayer((batch_size, 1, 32, 32), (neure[0], 1, 3, 3), 'relu',
'Gaussian', 0.0001))
Test_Single.add(PoolingLayer())
Test_Single.add(
ConvolutionLayer((batch_size, neure[0], 15, 15),
(neure[1], neure[0], 4, 4), 'relu', 'Gaussian', 0.01))
Test_Single.add(PoolingLayer())
Test_Single.add(
print "facial expression %d: test result %d/%d" %(y,ans,examples)
print " error rate %f%%" %(float(examples-ans)/examples*100)
global test_hit,test_count;
test_hit+=ans
test_count+=examples
def test_knn_stat():
print "\nNow tesing with KNN-ROI.....\n"
global test_hit,test_count;
for idx in xrange(5):
test_knn(("dataset/ROI_TEST/%d.bin")%idx,"dataset/ROI/mean.pkl",idx)
print "total error rate: %f%%\n"%(float(test_count-test_hit)/test_count*100)
# clear
test_hit=test_count=0
if __name__ == '__main__':
Test_Single=Model(batch_size=1,lr=0.01,dataSet=None)
meta_num=100
neure=[meta_num,meta_num,meta_num,meta_num]
batch_size=1
x=T.matrix('x')
index=T.lscalar()
Test_Single.add(DataLayer(batch_size,32*32))
Test_Single.add(FullyConnectedLayer(32*32,neure[0],'relu','Gaussian',0.1))
Test_Single.add(DropoutLayer(0.2))
Test_Single.add(FullyConnectedLayer(neure[0],neure[1],'relu','Gaussian',0.1))
Test_Single.add(DropoutLayer(0.2))
Test_Single.add(FullyConnectedLayer(neure[1],neure[2],'relu','Gaussian',0.1))
Test_Single.add(DropoutLayer(0.2))
Test_Single.add(SoftmaxLayer(neure[2],5))
Test_Single.build_test_fn()
Test_Single.load_params('params/DNN2000_ROI.pkl')
def test_knn_stat():
print "\nNow tesing with KNN-ROI.....\n"
global test_hit, test_count
for idx in xrange(5):
test_knn(("dataset/ROI_TEST/%d.bin") % idx, "dataset/ROI/mean.pkl",
idx)
print "total error rate: %f%%\n" % (float(test_count - test_hit) /
test_count * 100)
# clear
test_hit = test_count = 0
if __name__ == '__main__':
Test_Single = Model(batch_size=1, lr=0.01, dataSet=None)
meta_num = 100
neure = [meta_num, meta_num, meta_num, meta_num]
batch_size = 1
x = T.matrix('x')
index = T.lscalar()
Test_Single.add(DataLayer(batch_size, 32 * 32))
Test_Single.add(
FullyConnectedLayer(32 * 32, neure[0], 'relu', 'Gaussian', 0.1))
Test_Single.add(DropoutLayer(0.2))
Test_Single.add(SoftmaxLayer(neure[0], 5))
Test_Single.build_test_fn()
Test_Single.load_params('params/1NN2000_ROI.pkl')
test_pred = Test_Single.test_pred
test_belief = Test_Single.test_belief
test_single_stat()
from layer.core import *
from algorithm.SGD import Mini_Batch
from data.process import loadTrainData
from layer.model import Model
if __name__ == '__main__':
# load ROI+ROTATION dataset
dataSet = loadTrainData("dataset/ROT/data.pkl", "dataset/ROT/mean.pkl")
# load ROI dataset
#dataSet=loadTrainData("dataset/ROI/data.pkl","dataset/ROI/mean.pkl");
# load normal dataset
#dataSet=loadTrainData("dataset/NORMAL/data.pkl","dataset/NORMAL/mean.pkl");
cifar = Model(batch_size=100, lr=0.0001, dataSet=dataSet, weight_decay=0)
neure = [64, 64, 128, 300]
#neure=[32,32,64,64]
#neure=[48,48,96,200]
batch_size = 100
cifar.add(DataLayer(batch_size, (32, 32, 1)))
cifar.add(
ConvolutionLayer((batch_size, 1, 32, 32), (neure[0], 1, 3, 3), 'relu',
'Gaussian', 0.0001))
cifar.add(PoolingLayer())
cifar.add(
ConvolutionLayer((batch_size, neure[0], 15, 15),
(neure[1], neure[0], 4, 4), 'relu', 'Gaussian', 0.01))
cifar.add(PoolingLayer())
cifar.add(
ConvolutionLayer((batch_size, neure[1], 6, 6),
(neure[2], neure[1], 5, 5), 'relu', 'Gaussian', 0.01))
cifar.add(PoolingLayer())
cifar.add(
FullyConnectedLayer(neure[2] * 1 * 1, neure[3], 'relu', 'Gaussian',
print "facial expression %d: test result %d/%d" %(y,ans,examples)
print " error rate %f%%" %(float(examples-ans)/examples*100)
global test_hit,test_count;
test_hit+=ans
test_count+=examples
def test_knn_stat():
print "\nNow tesing with KNN-ROI.....\n"
global test_hit,test_count;
for idx in xrange(5):
test_knn(("dataset/ROI_TEST/%d.bin")%idx,"dataset/ROI/mean.pkl",idx)
print "total error rate: %f%%\n"%(float(test_count-test_hit)/test_count*100)
# clear
test_hit=test_count=0
if __name__ == '__main__':
Test_Single=Model(batch_size=1,lr=0.01,dataSet=None)
meta_num=100
neure=[meta_num,meta_num,meta_num,meta_num]
batch_size=1
x=T.matrix('x')
index=T.lscalar()
Test_Single.add(DataLayer(batch_size,(32,32,1)))
Test_Single.add(ConvolutionLayer((batch_size,1,32,32),(neure[0],1,3,3),'relu','Gaussian',0.0001))
Test_Single.add(PoolingLayer())
Test_Single.add(ConvolutionLayer((batch_size,neure[0],15,15),(neure[1],neure[0],4,4),'relu','Gaussian',0.01))
Test_Single.add(PoolingLayer())
Test_Single.add(ConvolutionLayer((batch_size,neure[1],6,6),(neure[2],neure[1],5,5),'relu','Gaussian',0.01))
Test_Single.add(PoolingLayer())
Test_Single.add(FullyConnectedLayer(neure[2]*1*1,neure[3],'relu','Gaussian',0.1))
Test_Single.add(DropoutLayer(0.5))
Test_Single.add(SoftmaxLayer(neure[3],5,'Gaussian',0.1))
from layer.core import *
from algorithm.SGD import Mini_Batch
from data.process import loadData, loadScaleData
from layer.model import Model
if __name__ == '__main__':
dataSet=loadScaleData('data.pkl')
cifar=Model(batch_size=100,lr=0.005,dataSet=dataSet,weight_decay=0.0)
neure=[1000,1000,1000]
batch_size=100
cifar.add(DataLayer(batch_size,32*32*3))
cifar.add(AutoEncodeLayer(32*32*3,neure[0],'relu','softplus',cost='squre',weight_init='Gaussian',gauss_std=0.1,level=0.3))
cifar.add(DropoutLayer(0.2))
cifar.add(SoftmaxLayer(neure[0],10))
cifar.pretrain(batch_size=20,n_epoches=15)
cifar.build_train_fn()
cifar.build_vaild_fn()
algorithm=Mini_Batch(model=cifar,n_epochs=100,load_param='',save_param='mlp_params.pkl')
algorithm.run()
