def single_run(te):
print te
data, label = make_moons(n_samples=2000,
noise=0.05,
shuffle=True,
random_state=int(time.time()))
data, validation_data, label, validation_label = train_test_split(
data, label, train_size=.30)
#separate the data set into buckets
total_data = []
for item in data: #list(group_list(data,1))
total_data.append(np.array(np.matrix(item)))
total_label = []
for item in label: #list(group_list(label,1))
total_label.append(np.asarray(item))
#The two separate site sets
for s in range(10, 150, 10):
nets = []
nn_groups_data = []
nn_groups_label = []
number_of_nets = s
for x in range(number_of_nets):
nets.append(nnDif.nn_build(1, [2, 6, 6, 1], eta=eta,
nonlin=nonlin))
iters = 100
for j in range(number_of_nets):
x = (total_data[int(float(j) / number_of_nets *
(len(total_data))):int(
float((j + 1)) / number_of_nets *
(len(total_data)))])
nn_groups_data.append(x)
nn_groups_label.append(total_label[int(
float(j) / number_of_nets * (len(total_label) / number_of_nets)
):int(float((j + 1)) / number_of_nets * (len(total_label)))])
start = time.time()
visitbatches(nets, nn_groups_data, nn_groups_label, [], it=iters)
one = accuracy(nets[0], validation_data, validation_label, thr=0.5)
nn1Acc[te][s / 10] += one
print "accuracy"
print one
'''
def single_run(te):
print te
data, label = make_moons(n_samples=2000, noise=0.05, shuffle=True, random_state = int(time.time()))
data,validation_data,label,validation_label = train_test_split(data,label,train_size = .30)
#separate the data set into buckets
total_data = []
for item in data:#list(group_list(data,1))
total_data.append(np.array(np.matrix(item)))
total_label = []
for item in label:#list(group_list(label,1))
total_label.append(np.asarray(item))
#The two separate site sets
for s in range(10,150,10):
nets = []
nn_groups_data = []
nn_groups_label = []
number_of_nets = s
for x in range(number_of_nets):
nets.append(nnDif.nn_build(1,[2,6,6,1],eta=eta,nonlin=nonlin))
iters = 100
for j in range(number_of_nets):
x = (total_data[int(float(j)/number_of_nets*(len(total_data))):int(float((j+1))/number_of_nets*(len(total_data)))])
nn_groups_data.append(x)
nn_groups_label.append(total_label[int(float(j)/number_of_nets*(len(total_label)/number_of_nets)):int(float((j+1))/number_of_nets*(len(total_label)))])
start = time.time()
visitbatches(nets,nn_groups_data,nn_groups_label,[],it=iters)
one = accuracy(nets[0], validation_data, validation_label, thr=0.5)
nn1Acc[te][s/10] += one
print "accuracy"
print one
'''
batches = list() #a list to store every separate site set
#Lists for our error to be plotted later
nat = []
for i in range((min(len(nn1_groups_data),len(nn2_groups_data),len(nn2_groups_data)))):#
print len(nn1_groups_data)-1
groups_data = list()
groups_label = list()
nets = list()
batches = list()
nnClassic1 = nnS.nn_build(1,[2,6,6,1],eta=eta,nonlin=nonlin)
nnClassic2 = nnS.nn_build(1,[2,6,6,1],eta=eta,nonlin=nonlin)
for x in range(number_of_nets):
nets.append(nnDif.nn_build(1,[2,6,6,1],eta=eta,nonlin=nonlin))
#Build the two site data sets
groups_data.append(np.asarray([item for sublist in nn1_groups_data[:i+1] for item in sublist]))
groups_data.append(np.asarray([item for sublist in nn2_groups_data[:i+1] for item in sublist]))
groups_data.append(np.asarray([item for sublist in nn3_groups_data[:i+1] for item in sublist]))
print "groups data length " + str(len(nn1_groups_label))
print "groups data length " + str(len(nn1_groups_data[i]))
print "groups data length " + str(len(nn2_groups_label[i]))
print "groups data length " + str(len(nn2_groups_data))
groups_label.append(np.asarray([item for sublist in nn1_groups_label[:i+1] for item in sublist]))
groups_label.append(np.asarray([item for sublist in nn2_groups_label[:i+1] for item in sublist]))
groups_label.append(np.asarray([item for sublist in nn3_groups_label[:i+1] for item in sublist]))
total_groups_data = np.asarray([item for sublist in groups_data for item in sublist])
total_groups_label = np.asarray([item for sublist in groups_label for item in sublist])
def sing_run(counts):
siteCount = 0
if counts[1] == 2:
siteCount = 0
if counts[1] == 10:
siteCount = 1
if counts[1] == 100:
siteCount = 2
number_of_nets = counts[1]
data, label = make_moons(n_samples=6 * site_size * number_of_nets, noise=0.05, shuffle=True, random_state = int(time.time()))
data,validation_data,label,validation_label = train_test_split(data,label,train_size = .50)
biases = [.01,.02,.05,.15,.20,.25,.30,.35,.40,.5]
#biases = [.01,.5]
biasCount = 0
for b in biases:
#nData, nLabel = (bias(data,label,b))
total_data = data
#print len(total_data)
total_label = label#list(group_list(nLabel,1))
nn_groups_data = []
nn_groups_label = list()
groups_data = list()
groups_label = list()
nets = list()
batches = list()
for j in range(number_of_nets):
dataCount = len(total_data) / number_of_nets
#x = (total_data[int(float(j)/number_of_nets*(len(total_data))):int(float((j+1))/number_of_nets*(len(total_data)))])
x = total_data[site_size * j * 3 : site_size*(j+1)*3]
y = total_label[site_size * j*3 : site_size*(j+1)*3]
d,l = bias(x,y,b,site_size)
nn_groups_data.append(d)
#print len(nn_groups_data[j])
#print "HERE BREAK"
#print str(float((j+1))) + " " + str(number_of_nets) + " LEN TOTAL DATA: " + str(len(total_data))
#print int(float((j+1))/number_of_nets*(len(total_data)/number_of_nets))
#print "HERE END"
nn_groups_label.append(l)
#print len(nn_groups_data[1])
nets = list() #Our differential networks
batches = list() #a list to store every separate site set
labelBatches = list()
#Lists for our error to be plotted later
#FIXME: Needs to fill batches only. Really though? Think about this more...
#TODO: run this guy when you get the chance. 10 samples, and check. if it's not what you want,
#it's a problem...
#for i in range(len(nn_groups_data)):
#nnClassic1 = nnS.nn_build(1,[2,6,6,1],eta=eta,nonlin=nonlin)
for x in range(number_of_nets):
nets.append(nnDif.nn_build(1,[2,6,6,1],eta=eta,nonlin=nonlin))
groups_data = []
groups_label = []
print "y"
for ke in range(0,number_of_nets):
for k in range(0,site_size):
groups_data.append((nn_groups_data[ke][k].T))
groups_label.append(nn_groups_label[ke][k])
#t = [x for x in iter_minibatches(2,total_groups_data.T, total_groups_label)]
#t3 = [x for x in iter_minibatches(2,total_groups_data.T, total_groups_label)]
err = []
#Run the batches through the algos
iters = 100
#visitClassicBatches(nnClassic1,t, it=iters)
#visitClassicBatches(nnClassic3,t3, it=iters)
start = time.time()
batches_data,batches_label = [x for x in iter_minibatches(site_size,groups_data, groups_label)]
visitClassicBatches(nets[0], batches_data, batches_label, it=iters)
#print time.time() - start
#calculate error
#classic = accuracyClassic(nnClassic1,validation_data,validation_label, thr=0.5)
print "has finished calc"
one = accuracy(nets[0], validation_data, validation_label, thr=0.5)
#classic3 = accuracyClassic(nnClassic3,validation_data,validation_label, thr=0.5)
nat = nets[0]
#build plottable arrays
nn1Acc[siteCount][counts[0]][biasCount] += one
print "accuracy"
print one
#classAcc1[te][s/2] += classic
#classAcc3[te][s/2] += classic3
#print "us " + str(one) + " c1 " + str(classic) + " cc " + str(classic3)
biasCount +=1
nets = list() #Our differential networks
batches = list() #a list to store every separate site set
for i in range((min(len(nn1_groups_data), len(nn2_groups_data),
len(nn2_groups_data)))): #
print len(nn1_groups_data) - 1
groups_data = list()
groups_label = list()
nets = list()
batches = list()
nnClassic1 = nnS.nn_build(1, [2, 6, 6, 1], eta=eta, nonlin=nonlin)
nnClassic2 = nnS.nn_build(1, [2, 6, 6, 1], eta=eta, nonlin=nonlin)
for x in range(number_of_nets):
nets.append(nnDif.nn_build(1, [2, 6, 6, 1], eta=eta,
nonlin=nonlin))
#Build the two site data sets
groups_data.append(
np.asarray([
item for sublist in nn1_groups_data[:i + 1] for item in sublist
]))
groups_data.append(
np.asarray([
item for sublist in nn2_groups_data[:i + 1] for item in sublist
]))
groups_data.append(
np.asarray([
item for sublist in nn3_groups_data[:i + 1] for item in sublist
]))
print "groups data length " + str(len(nn1_groups_label))
def sing_run(te):
print te
data, label = make_moons(n_samples=2000, shuffle=True, noise=0.1,random_state = int(time.time()))
data,validation_data,label,validation_label = train_test_split(data,label,train_size = .38)
#separate the data set into buckets
total_data, total_label,A,B = split(data,label)
random.seed(4)
random.shuffle(total_data[0])
random.seed(4)
random.shuffle(total_data[1])
random.seed(4)
random.shuffle(total_data[2])
random.seed(4)
random.shuffle(total_label[0])
random.seed(4)
random.shuffle(total_label[1])
random.seed(4)
random.shuffle(total_label[2])
totlistX = []
totlistY = []
#for item in total_data[0]:
#x=2
#totlistX.append(item[0])
#totlistY.append(item[1])
#for item in total_data[1]:
#x=2
#totlistX.append(item[0])
#totlistY.append(item[1])
#for item in total_data[2]:
#x=2
#totlistX.append(item[0])
#totlistY.append(item[1])
# print len(totlistX)
#plt.scatter(totlistX,totlistY,c=total_label[0] + total_label[1] +total_label[2])
#+total_label[1]+total_label[2]
#random.shuffle(total_label[2])
'''for i in range(3):
for item in total_data[i]:
for j in range(len(item)):
total_data[i][j] = list(total_data[i][j])'''
#total_data = list(group_list(data,10))
#total_label = list(group_list(label,10))
#The two separate site sets
#nn1_groups_data = total_data[:len(total_data)/2+1]
#nn1_groups_data = total_data[0]#list(group_list(total_data[0],1))
#nn1_groups_label = total_label[0]#list(group_list(total_label[0],1))
#nn2_groups_data = total_data[1]#list(group_list(total_data[1],1))
#nn2_groups_label = total_label[1]#list(group_list(total_label[1],1))
#nn3_groups_data = total_data[2]#list(group_list(total_data[2],1))
#nn3_groups_label = total_label[2]#list(group_list(total_label[2],1))
minim = min(min(len(total_data[0]),len(total_data[1])),len(total_data[2]))
#print "length of site one group data " + str(len(nn1_groups_data))
#nn2_groups_data = total_data[len(total_data)/2:]
#nn1_groups_label = total_label[:len(total_data)/2+1]
#nn2_groups_label = total_label[len(total_data)/2:]
nets = list() #Our differential networks
batches = list() #a list to store every separate site set
#Lists for our error to be plotted later
nat = []
for i in range((minim - minim%10) -10,minim - minim%10,10):#
totlistX1 = []
totlistY1 = []
totlistX2 = []
totlistY2 = []
totlistX3 = []
totlistY3 = []
for item in total_data[0][:i]:
x=2
totlistX1.append(item[0])
totlistY1.append(item[1])
for item in total_data[1][:i]:
x=2
totlistX1.append(item[0])
totlistY1.append(item[1])
for item in total_data[2][:i]:
totlistX1.append(item[0])
totlistY1.append(item[1])
#print len(totlistX)
#print len(total_label[0][:i]+total_label[1][:i]+total_label[2][:i])
#print "y"
#print [0 for x in total_label[0][:i]]+[1 for x in total_label[1][:i]]+[2 for x in total_label[2][:i]]
#plt.scatter(totlistX1,totlistY1,c=[0 for x in total_label[0][:i]]+[1 for x in total_label[1][:i]]+[2 for x in total_label[2][:i]])
#plt.scatter(totlistX2,totlistY2,c=[x + 2 for x in total_label[1][:i]])
#plt.scatter(totlistX3,totlistY3,c=[x + 4 for x in total_label[2][:i]])
#plt.show()
#exit(1)
#TODO: Here, we need to rewrite the function so it
groups_data = []
groups_label = []
nets = []
batches = []
#nnClassic1 = nnS.nn_build(1,[2,6,6,1],eta=eta,nonlin=nonlin)
#nnClassic2 = nnS.nn_build(1,[2,6,6,1],eta=eta,nonlin=nonlin)
for x in range(number_of_nets):
nets.append(nnDif.nn_build(1,[2,6,6,1],eta=eta,nonlin=nonlin))
print number_of_nets
nextNet = nnDif.nn_build(1,[2,6,6,1],eta=eta,nonlin=nonlin)
#Build the two site data sets
#groups_data.append(np.asarray([item for sublist in nn1_groups_data[:i+1] for item in sublist]))
#groups_data.append(np.asarray([item for sublist in nn2_groups_data[:i+1] for item in sublist]))
#groups_data.append(np.asarray([item for sublist in nn3_groups_data[:i+1] for item in sublist]))
#groups_label.append(np.asarray([item for sublist in nn1_groups_label[:i+1] for item in sublist]))
#groups_label.append(np.asarray([item for sublist in nn2_groups_label[:i+1] for item in sublist]))
#groups_label.append(np.asarray([item for sublist in nn3_groups_label[:i+1] for item in sublist]))
#total_groups_data = np.asarray([item for sublist in groups_data for item in sublist])
#total_groups_label = np.asarray([item for sublist in groups_label for item in sublist])
#Our classic combined nn
#nnClassic3 = nnS.nn_build(1,[2,6,6,1],eta=eta,nonlin=nonlin)
#nnClassic4 = nnS.nn_build(1,[2,6,6,1],eta=eta,nonlin=nonlin)
#for s in range(number_of_nets):
#batches.append([x for x in iter_minibatches(2,groups_data[s].T, groups_label[s])])
#t = [x for x in iter_minibatches(2,groups_data[0].T, groups_label[0])]
#t2 = [x for x in iter_minibatches(1,groups_data[1].T, groups_label[1])]
#t3 = [x for x in iter_minibatches(1,groups_data[2].T, groups_label[2])]
#t4 = [x for x in iter_minibatches(1,total_groups_data.T, total_groups_label)]
err = []
#Run the batches through the algos
iters = 700
#visitClassicBatches(nnClassic2,nn2_groups_data[:i],nn2_groups_label[:i], it=iters)
#visitClassicBatches(nnClassic3,nn3_groups_data[:i],nn3_groups_label[:i], it=iters)
#visitClassicBatches(nnClassic4,t4, it=iters)
#visitbatches(nets, batches, err, it=iters)
#print "finish classics"
#differential_groups = differential_groups + dif_group_data[3*i] +dif_group_data[3*i + 1] + dif_group_data[3*i + 2]
batchesData1,batchesLabel1 = [x for x in iter_minibatches(1,total_data[0],total_label[0])]
batchesData2,batchesLabel2 = [x for x in iter_minibatches(1,total_data[1],total_label[1])]
batchesData3,batchesLabel3 = [x for x in iter_minibatches(1,total_data[2],total_label[2])]
nnTogetherClassic = nnS.nn_build(1,[2,6,6,1],eta=eta,nonlin=nonlin)
classicWing = nnS.nn_build(1,[2,6,6,1],eta=eta,nonlin=nonlin)
classicWing2 = nnS.nn_build(1,[2,6,6,1],eta=eta,nonlin=nonlin)
classicLeft = nnS.nn_build(1,[2,6,6,1],eta=eta,nonlin=nonlin)
nnTogetherDif = nnDif.nn_build(1,[2,6,6,1],eta=eta,nonlin=nonlin)
print "LENGTH: " +str(len(batchesData1[:i]))
visitClassicBatches(classicWing,batchesData2[:i],batchesLabel2[:i],it=iters)
#visitClassicBatches(classicWing2,batchesData2[:i],batchesLabel2[:i],it=iters)
visitClassicBatches(classicLeft,batchesData1[:i],batchesLabel1[:i],it=iters)
visitClassicBatches(nnTogetherClassic,batchesData1[:i]+batchesData2[:i]+batchesData3[:i],batchesLabel1[:i]+batchesLabel2[:i]+batchesLabel3[:i],it=iters)
#visitClassicBatches(nnClassic1,batchesData1[:i],batchesLabel1[:i], it=iters)
#visitClassicBatches(nnClassic2,batchesData2[:i],batchesLabel2[:i], it=iters)
#visitClassicBatches(nnClassic3,batchesData3[:i],batchesLabel3[:i], it=iters)
plotlistX = []
plotlistY = []
plotlistColor = []
for batch in batchesData1[:i]:
for tup in batch:
plotlistX.append(tup[0])
plotlistY.append(tup[1])
#for batch in batchesData2[:i]:
# for tup in batch:
# for sid in tup:
# plotlistX.append(sid[0])
# plotlistY.append(sid[1])
#for batch in batchesData3[:i]:
# for tup in batch:
# x = 2
#plotlistX.append(tup[0])
#plotlistY.append(tup[1])
for arr in batchesLabel1[:i]:
for tup in arr:
plotlistColor.append(tup)
#for arr in batchesLabel2[:i]:
# for tup in arr:
# plotlistColor.append(tup[0])
# plotlistColor.append(tup[0])
#plt.scatter(plotlistX,plotlistY,c=plotlistColor)
#+batchesLabel2[:i]+batchesLabel3[:i])
#plt.show()
new_total_data = []
new_total_label = []
for x in range(i):
for se in range(len(nets)):
new_total_data.append(total_data[0][x])
new_total_data.append(total_data[1][x])
new_total_data.append(total_data[2][x])
new_total_label.append(total_label[0][x])
new_total_label.append(total_label[1][x])
new_total_label.append(total_label[2][x])
#differential_labels = differential_labels + dif_group_label[3*i] + dif_group_label[3*i + 1] + dif_group_label[3*i + 2]
#visitbatches(nets, [nn1_groups_data[:i],nn2_groups_data[:i],nn3_groups_data[:i]], [nn1_groups_label[:i],nn2_groups_label[:i],nn3_groups_label[:i]], err, it=iters)
batchesList = [batchesData1[:i],batchesData2[:i],batchesData3[:i]]
labelList = [batchesLabel1[:i],batchesLabel2[:i],batchesLabel3[:i]]
visitClassicBatches(nets[0], [batchesData1[:i],batchesData2[:i],batchesData3[:i]], [batchesLabel1[:i],batchesLabel2[:i],batchesLabel3[:i]], err, it=iters)
batches2 = [batchesData1[:i]+batchesData2[:i]+batchesData3[:i]]
list2 = [batchesLabel1[:i]+batchesLabel2[:i]+batchesLabel3[:i]]
#visitbatches([nets[1]], [batchesData1[:i]], [batchesLabel1[:i]], err, it=iters)
#visitbatches([nextNet], [batchesData1[:i]+batchesData2[:i]+batchesData3[:i]], [batchesLabel1[:i]+batchesLabel2[:i]+batchesLabel3[:i]], err, it=iters)
#calculate error
#nnClassic5 = nnS.nn_build(1,[2,6,6,1],eta=eta,nonlin=nonlin)
#testAcc = accuracyClassic(nnClassic5,validation_data,validation_label,thr=.05)
#testAcc2 = accuracy(nnClassic5,validation_data,validation_label,thr=.05)
togetherAcc = accuracyClassic(nnTogetherClassic,validation_data,validation_label,thr=.05)
classic = accuracyClassic(classicWing,validation_data,validation_label, thr=0.5)
#classic2 = accuracyClassic(classicWing2,validation_data,validation_label, thr=0.5)
classicLeft = accuracyClassic(classicLeft,validation_data,validation_label, thr=0.5)
one = accuracyClassic(nets[0], validation_data, validation_label, thr=0.5)
#two = accuracy(nets[1], validation_data,validation_label, thr=0.5)
#classic2 = accuracyClassic(nnClassic2,validation_data,validation_label, thr=0.5)
#classic3 = accuracyClassic(nnClassic3,validation_data,validation_label, thr=0.5)
#classic4 = accuracyClassic(nnClassic4,validation_data,validation_label, thr=0.5)
#build plottable arrays
nn1Acc[te][i/10] = one
#newAcc = accuracy(nnTogetherDif,validation_data,validation_label,thr=.5)
print "ACCURACY"
print one
print "^^^Decent acc || total acc"
print togetherAcc
print "MIDDLE " + str(classic)
print "Left side: "+ str(classicLeft)
classAcc1[te][i/10] = classic
#classAcc2[te][i/10] = classic2
classAcc3[te][i/10] = classicLeft
classAcc4[te][i/10] = togetherAcc
def sing_run(te):
print te
data, label = make_moons(n_samples=2000,
shuffle=True,
noise=0.1,
random_state=int(time.time()))
data, validation_data, label, validation_label = train_test_split(
data, label, train_size=.38)
#separate the data set into buckets
total_data, total_label, A, B = split(data, label)
random.seed(4)
random.shuffle(total_data[0])
random.seed(4)
random.shuffle(total_data[1])
random.seed(4)
random.shuffle(total_data[2])
random.seed(4)
random.shuffle(total_label[0])
random.seed(4)
random.shuffle(total_label[1])
random.seed(4)
random.shuffle(total_label[2])
totlistX = []
totlistY = []
#for item in total_data[0]:
#x=2
#totlistX.append(item[0])
#totlistY.append(item[1])
#for item in total_data[1]:
#x=2
#totlistX.append(item[0])
#totlistY.append(item[1])
#for item in total_data[2]:
#x=2
#totlistX.append(item[0])
#totlistY.append(item[1])
# print len(totlistX)
#plt.scatter(totlistX,totlistY,c=total_label[0] + total_label[1] +total_label[2])
#+total_label[1]+total_label[2]
#random.shuffle(total_label[2])
'''for i in range(3):
for item in total_data[i]:
for j in range(len(item)):
total_data[i][j] = list(total_data[i][j])'''
#total_data = list(group_list(data,10))
#total_label = list(group_list(label,10))
#The two separate site sets
#nn1_groups_data = total_data[:len(total_data)/2+1]
#nn1_groups_data = total_data[0]#list(group_list(total_data[0],1))
#nn1_groups_label = total_label[0]#list(group_list(total_label[0],1))
#nn2_groups_data = total_data[1]#list(group_list(total_data[1],1))
#nn2_groups_label = total_label[1]#list(group_list(total_label[1],1))
#nn3_groups_data = total_data[2]#list(group_list(total_data[2],1))
#nn3_groups_label = total_label[2]#list(group_list(total_label[2],1))
minim = min(min(len(total_data[0]), len(total_data[1])),
len(total_data[2]))
#print "length of site one group data " + str(len(nn1_groups_data))
#nn2_groups_data = total_data[len(total_data)/2:]
#nn1_groups_label = total_label[:len(total_data)/2+1]
#nn2_groups_label = total_label[len(total_data)/2:]
nets = list() #Our differential networks
batches = list() #a list to store every separate site set
#Lists for our error to be plotted later
nat = []
for i in range((minim - minim % 10) - 10, minim - minim % 10, 10): #
totlistX1 = []
totlistY1 = []
totlistX2 = []
totlistY2 = []
totlistX3 = []
totlistY3 = []
for item in total_data[0][:i]:
x = 2
totlistX1.append(item[0])
totlistY1.append(item[1])
for item in total_data[1][:i]:
x = 2
totlistX1.append(item[0])
totlistY1.append(item[1])
for item in total_data[2][:i]:
totlistX1.append(item[0])
totlistY1.append(item[1])
#print len(totlistX)
#print len(total_label[0][:i]+total_label[1][:i]+total_label[2][:i])
#print "y"
#print [0 for x in total_label[0][:i]]+[1 for x in total_label[1][:i]]+[2 for x in total_label[2][:i]]
#plt.scatter(totlistX1,totlistY1,c=[0 for x in total_label[0][:i]]+[1 for x in total_label[1][:i]]+[2 for x in total_label[2][:i]])
#plt.scatter(totlistX2,totlistY2,c=[x + 2 for x in total_label[1][:i]])
#plt.scatter(totlistX3,totlistY3,c=[x + 4 for x in total_label[2][:i]])
#plt.show()
#exit(1)
#TODO: Here, we need to rewrite the function so it
groups_data = []
groups_label = []
nets = []
batches = []
#nnClassic1 = nnS.nn_build(1,[2,6,6,1],eta=eta,nonlin=nonlin)
#nnClassic2 = nnS.nn_build(1,[2,6,6,1],eta=eta,nonlin=nonlin)
for x in range(number_of_nets):
nets.append(nnDif.nn_build(1, [2, 6, 6, 1], eta=eta,
nonlin=nonlin))
print number_of_nets
nextNet = nnDif.nn_build(1, [2, 6, 6, 1], eta=eta, nonlin=nonlin)
#Build the two site data sets
#groups_data.append(np.asarray([item for sublist in nn1_groups_data[:i+1] for item in sublist]))
#groups_data.append(np.asarray([item for sublist in nn2_groups_data[:i+1] for item in sublist]))
#groups_data.append(np.asarray([item for sublist in nn3_groups_data[:i+1] for item in sublist]))
#groups_label.append(np.asarray([item for sublist in nn1_groups_label[:i+1] for item in sublist]))
#groups_label.append(np.asarray([item for sublist in nn2_groups_label[:i+1] for item in sublist]))
#groups_label.append(np.asarray([item for sublist in nn3_groups_label[:i+1] for item in sublist]))
#total_groups_data = np.asarray([item for sublist in groups_data for item in sublist])
#total_groups_label = np.asarray([item for sublist in groups_label for item in sublist])
#Our classic combined nn
#nnClassic3 = nnS.nn_build(1,[2,6,6,1],eta=eta,nonlin=nonlin)
#nnClassic4 = nnS.nn_build(1,[2,6,6,1],eta=eta,nonlin=nonlin)
#for s in range(number_of_nets):
#batches.append([x for x in iter_minibatches(2,groups_data[s].T, groups_label[s])])
#t = [x for x in iter_minibatches(2,groups_data[0].T, groups_label[0])]
#t2 = [x for x in iter_minibatches(1,groups_data[1].T, groups_label[1])]
#t3 = [x for x in iter_minibatches(1,groups_data[2].T, groups_label[2])]
#t4 = [x for x in iter_minibatches(1,total_groups_data.T, total_groups_label)]
err = []
#Run the batches through the algos
iters = 700
#visitClassicBatches(nnClassic2,nn2_groups_data[:i],nn2_groups_label[:i], it=iters)
#visitClassicBatches(nnClassic3,nn3_groups_data[:i],nn3_groups_label[:i], it=iters)
#visitClassicBatches(nnClassic4,t4, it=iters)
#visitbatches(nets, batches, err, it=iters)
#print "finish classics"
#differential_groups = differential_groups + dif_group_data[3*i] +dif_group_data[3*i + 1] + dif_group_data[3*i + 2]
batchesData1, batchesLabel1 = [
x for x in iter_minibatches(1, total_data[0], total_label[0])
]
batchesData2, batchesLabel2 = [
x for x in iter_minibatches(1, total_data[1], total_label[1])
]
batchesData3, batchesLabel3 = [
x for x in iter_minibatches(1, total_data[2], total_label[2])
]
nnTogetherClassic = nnS.nn_build(1, [2, 6, 6, 1],
eta=eta,
nonlin=nonlin)
classicWing = nnS.nn_build(1, [2, 6, 6, 1], eta=eta, nonlin=nonlin)
classicWing2 = nnS.nn_build(1, [2, 6, 6, 1], eta=eta, nonlin=nonlin)
classicLeft = nnS.nn_build(1, [2, 6, 6, 1], eta=eta, nonlin=nonlin)
nnTogetherDif = nnDif.nn_build(1, [2, 6, 6, 1], eta=eta, nonlin=nonlin)
print "LENGTH: " + str(len(batchesData1[:i]))
visitClassicBatches(classicWing,
batchesData2[:i],
batchesLabel2[:i],
it=iters)
#visitClassicBatches(classicWing2,batchesData2[:i],batchesLabel2[:i],it=iters)
visitClassicBatches(classicLeft,
batchesData1[:i],
batchesLabel1[:i],
it=iters)
visitClassicBatches(
nnTogetherClassic,
batchesData1[:i] + batchesData2[:i] + batchesData3[:i],
batchesLabel1[:i] + batchesLabel2[:i] + batchesLabel3[:i],
it=iters)
#visitClassicBatches(nnClassic1,batchesData1[:i],batchesLabel1[:i], it=iters)
#visitClassicBatches(nnClassic2,batchesData2[:i],batchesLabel2[:i], it=iters)
#visitClassicBatches(nnClassic3,batchesData3[:i],batchesLabel3[:i], it=iters)
plotlistX = []
plotlistY = []
plotlistColor = []
for batch in batchesData1[:i]:
for tup in batch:
plotlistX.append(tup[0])
plotlistY.append(tup[1])
#for batch in batchesData2[:i]:
# for tup in batch:
# for sid in tup:
# plotlistX.append(sid[0])
# plotlistY.append(sid[1])
#for batch in batchesData3[:i]:
# for tup in batch:
# x = 2
#plotlistX.append(tup[0])
#plotlistY.append(tup[1])
for arr in batchesLabel1[:i]:
for tup in arr:
plotlistColor.append(tup)
#for arr in batchesLabel2[:i]:
# for tup in arr:
# plotlistColor.append(tup[0])
# plotlistColor.append(tup[0])
#plt.scatter(plotlistX,plotlistY,c=plotlistColor)
#+batchesLabel2[:i]+batchesLabel3[:i])
#plt.show()
new_total_data = []
new_total_label = []
for x in range(i):
for se in range(len(nets)):
new_total_data.append(total_data[0][x])
new_total_data.append(total_data[1][x])
new_total_data.append(total_data[2][x])
new_total_label.append(total_label[0][x])
new_total_label.append(total_label[1][x])
new_total_label.append(total_label[2][x])
#differential_labels = differential_labels + dif_group_label[3*i] + dif_group_label[3*i + 1] + dif_group_label[3*i + 2]
#visitbatches(nets, [nn1_groups_data[:i],nn2_groups_data[:i],nn3_groups_data[:i]], [nn1_groups_label[:i],nn2_groups_label[:i],nn3_groups_label[:i]], err, it=iters)
batchesList = [batchesData1[:i], batchesData2[:i], batchesData3[:i]]
labelList = [batchesLabel1[:i], batchesLabel2[:i], batchesLabel3[:i]]
visitClassicBatches(
nets[0], [batchesData1[:i], batchesData2[:i], batchesData3[:i]],
[batchesLabel1[:i], batchesLabel2[:i], batchesLabel3[:i]],
err,
it=iters)
batches2 = [batchesData1[:i] + batchesData2[:i] + batchesData3[:i]]
list2 = [batchesLabel1[:i] + batchesLabel2[:i] + batchesLabel3[:i]]
#visitbatches([nets[1]], [batchesData1[:i]], [batchesLabel1[:i]], err, it=iters)
#visitbatches([nextNet], [batchesData1[:i]+batchesData2[:i]+batchesData3[:i]], [batchesLabel1[:i]+batchesLabel2[:i]+batchesLabel3[:i]], err, it=iters)
#calculate error
#nnClassic5 = nnS.nn_build(1,[2,6,6,1],eta=eta,nonlin=nonlin)
#testAcc = accuracyClassic(nnClassic5,validation_data,validation_label,thr=.05)
#testAcc2 = accuracy(nnClassic5,validation_data,validation_label,thr=.05)
togetherAcc = accuracyClassic(nnTogetherClassic,
validation_data,
validation_label,
thr=.05)
classic = accuracyClassic(classicWing,
validation_data,
validation_label,
thr=0.5)
#classic2 = accuracyClassic(classicWing2,validation_data,validation_label, thr=0.5)
classicLeft = accuracyClassic(classicLeft,
validation_data,
validation_label,
thr=0.5)
one = accuracyClassic(nets[0],
validation_data,
validation_label,
thr=0.5)
#two = accuracy(nets[1], validation_data,validation_label, thr=0.5)
#classic2 = accuracyClassic(nnClassic2,validation_data,validation_label, thr=0.5)
#classic3 = accuracyClassic(nnClassic3,validation_data,validation_label, thr=0.5)
#classic4 = accuracyClassic(nnClassic4,validation_data,validation_label, thr=0.5)
#build plottable arrays
nn1Acc[te][i / 10] = one
#newAcc = accuracy(nnTogetherDif,validation_data,validation_label,thr=.5)
print "ACCURACY"
print one
print "^^^Decent acc || total acc"
print togetherAcc
print "MIDDLE " + str(classic)
print "Left side: " + str(classicLeft)
classAcc1[te][i / 10] = classic
#classAcc2[te][i/10] = classic2
classAcc3[te][i / 10] = classicLeft
classAcc4[te][i / 10] = togetherAcc