def main():
# Import data
dic = cPickle.load(open("cifar_2class_py2.p","rb"))
train_data = (dic['train_data'] - dic['train_data'].mean(0)) / dic['train_data'].std(0)
train_data.shape = (10000, 3072)
#train_data = dic['train_data'] / 255
train_label = dic['train_labels']
test_data = (dic['test_data'] - dic['test_data'].mean(0)) / dic['test_data'].std(0)
test_data.shape = (2000, 3072)
#test_data = dic['test_data'] / 255
test_label = dic['test_labels']
# Create new net
net = MyOneLayerFullyConnectedNet(train_data, train_label, test_data, test_label)
for stepSIze in range(0.001, 1, 0.002):
for gamma in
# Create layer-level parameter
# Parameter format:
# [num, activation function, derivative of activation function]
specification = ((train_data.shape[1], 0, 0), (10, relu, relu_prime), (1, expit, sigmoid_prime))
# Start calculation
net.mySDGwithMomentum(param = specification, loss = crossEntropy, lossPrime = corssEngtropyPrime)
def main():
# Import data
dic = cPickle.load(open("cifar_2class_py2.p","rb"))
train_data = (dic['train_data'] - dic['train_data'].mean(0)) / dic['train_data'].std(0)
train_data.shape = (10000, 3072)
#train_data = dic['train_data'] / 255
train_label = dic['train_labels']
test_data = (dic['test_data'] - dic['test_data'].mean(0)) / dic['test_data'].std(0)
test_data.shape = (2000, 3072)
#test_data = dic['test_data'] / 255
test_label = dic['test_labels']
# Create new net
net = MyOneLayerFullyConnectedNet(train_data, train_label, test_data, test_label)
# parameters
stepSizes = [0.001, 0.003, 0.005, 0.007, 0.009, 0.011, 0.013, 0.015, 0.017, 0.019]
gammas = [0.6, 0.7, 0.8, 0.9]
bestStepSize = 0
bestBatchSize = 0
bestNumOfNeurons = 0
bestGamma = 0
curAcc = 0
globalOptimalAcc = 0
res = {}
# Tuning through stepSize, number of neuron, batch size;
for stepSizesIdx in range(10):
for numOfNeuron in range(10 , 101, 1):
for batchSize in range(32, 258, 2):
for gammaIdx in range(4):
print "Gamma:", gammas[gammaIdx]
# Create layer-level parameter
# Parameter format:
# [num, activation function, derivative of activation function]
specification = ((train_data.shape[1], 0, 0), (numOfNeuron, relu, relu_prime), (1, sigmoid, sigmoid_prime))
# Start calculation
curAcc = net.mySDGwithMomentum(param = specification, loss = crossEntropy, lossPrime = corssEngtropyPrime, miniBatchSize = batchSize, stepSize = stepSizes[stepSizesIdx], gamma = gammas[gammaIdx])
if curAcc > globalOptimalAcc:
bestStepSize = stepSizes[stepSizesIdx]
bestBatchSize = batchSize
bestNumOfNeurons = numOfNeuron
bestGamma = gammas[gammaIdx]
globalOptimalAcc = curAcc
#end
print "\n \n \n", "Current Acc: ", curAcc, "\n \n \n"
res["bestStepSize"] = stepSizes[stepSizesIdx]
res["bestBatchSize"] = batchSize
res["bestNumOfNeurons"] = numOfNeuron
res["bestGamma"] = gammas[gammaIdx]
res["curAcc"] = curAcc
cPickle.dump(res, open("/nfs/stak/students/l/lix3/dl/try/myRes/global/" + "_" + str(stepSizesIdx) + "_" + str(batchSize) + "_" + str(numOfNeuron) + "_" + str(gammaIdx) +".p", "wb"))
# end
# end
# end
# end
print "Best step size: ", bestStepSize
print "Best batch size: ", bestBatchSize
print "Best number of neurons: ", bestNumOfNeurons
print "Best gamma(momentum): ", bestGamma
print "Best Acc:", globalOptimalAcc
