def main():
start = datetime.datetime.now()
if not os.path.exists(args.checkpoint_dir):
os.makedirs(args.checkpoint_dir)
if not os.path.exists(args.sample_dir):
os.makedirs(args.sample_dir)
if not os.path.exists(args.test_dir):
os.makedirs(args.test_dir)
if not os.path.exists(args.train_dir):
os.makedirs(args.train_dir)
tf.reset_default_graph()
with tf.Session() as sess:
model = graph2graph(sess,
batch_size=args.batch_size,
checkpoint_dir=args.checkpoint_dir,
sample_dir=args.sample_dir,
test_dir=args.test_dir,
train_dir=args.train_dir,
graph_size=args.graph_size,
output_size=args.output_size,
dataset=args.dataset)
model.train(args)
end = datetime.datetime.now()
print(end - start)
def main(_):
if not os.path.exists(args.checkpoint_dir):
os.makedirs(args.checkpoint_dir)
tf.reset_default_graph()
with tf.Session() as sess:
model = graph2graph(sess,
Ds=args.Ds,
No=args.No,
Nr=args.Nr,
Dr=args.Dr,
Dx=args.Dx,
De_o=args.De_o,
De_r=args.De_r,
Mini_batch=args.Mini_batch,
checkpoint_dir=args.checkpoint_dir,
epoch=args.epoch,
Ds_inter=args.Ds_inter,
Dr_inter=args.Dr_inter)
if args.Type == 'train':
model.train(args)
if args.Type == 'test':
start = datetime.datetime.now()
model.test(args)
end = datetime.datetime.now()
print(end - start)
def main_auth(filename):
#start = datetime.datetime.now()
if not os.path.exists(args.checkpoint_dir):
os.makedirs(args.checkpoint_dir)
if not os.path.exists(args.sample_dir):
os.makedirs(args.sample_dir)
if not os.path.exists(args.test_dir):
os.makedirs(args.test_dir)
tf.reset_default_graph()
with tf.Session() as sess:
model = graph2graph(sess,
batch_size=args.batch_size,
checkpoint_dir=args.checkpoint_dir,
sample_dir=args.sample_dir,
test_dir=args.test_dir,
train_dir=args.train_dir,
graph_size=args.graph_size,
output_size=args.output_size)
#model.train(args)
model.test_auth(args, filename)
def main():
start = datetime.datetime.now()
if not os.path.exists(args.checkpoint_dir):
os.makedirs(args.checkpoint_dir)
if not os.path.exists(args.sample_dir):
os.makedirs(args.sample_dir)
if not os.path.exists(args.test_dir):
os.makedirs(args.test_dir)
#with tf.Session() as sess:
#model = graph2graph(sess, batch_size=args.batch_size,checkpoint_dir=args.checkpoint_dir,sample_dir=args.sample_dir,test_dir=args.test_dir,train_dir=args.train_dir,graph_size=args.graph_size,output_size=args.output_size, dataset=args.dataset)
v1 = np.load('data1s.npy')
v4 = np.load('data2s.npy')
predicted = np.zeros(shape=(30, 1225)) # full predicted dataset
ground = np.zeros(shape=(
30, 1225
)) # structure for test data (unused for traınıng) ın each ıteratıon
#loo = KFold(n_splits=5)
#loo.get_n_splits(v1)
#KFold(n_splits=6, random_state=None, shuffle=False)
from sklearn.model_selection import KFold
loo = KFold(n_splits=5)
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
session = tf.Session(config=config)
option = None
#TF_CONFIG = tf.ConfigProto(gpu_options=tf.GPUOptions(per_process_gpu_memory_fraction=0.5),allow_soft_placement=True)
#sess = tf.Session(config=TF_CONFIG)
pcr = []
mser = []
pv = []
predicteddata = []
wholepredicted = []
for train_index, test_index in loo.split(v1):
tf.reset_default_graph()
with tf.Session() as sess:
model = graph2graph(sess,
batch_size=args.batch_size,
checkpoint_dir=args.checkpoint_dir,
sample_dir=args.sample_dir,
test_dir=args.test_dir,
train_dir=args.train_dir,
graph_size=args.graph_size,
output_size=args.output_size,
dataset=args.dataset)
#epoch=args.epoch
##cuda.select_device(0)
reg1, reg2 = v1[train_index], v1[
test_index] # reg1 ıs v1 traın dataset, reg2 v1 testdataset
print(np.shape(reg1), 'reg1')
print(np.shape(reg2), 'reg2')
mal1, mal2 = v4[train_index], v4[
test_index] #mal1:v4 traın dataset,mal2 v4 testdataset(lıke y traın and test
datatrain = np.zeros((reg1.shape[0], reg1.shape[1], reg1.shape[2],
2)) # forming a paır of traın samples
print('now the test index is', test_index)
for i in range(reg1.shape[0]):
datatrain[i, :, :, 0] = reg1[i] ###reg1=train v1
datatrain[i, :, :,
1] = mal1[i] #mal1=train v2 (154, 35, 35, 2)
print(np.shape(datatrain), 'thıs ıs traın data after loop')
#model.train(args, datatrain)
datatest = np.zeros((reg2.shape[0], reg2.shape[1], reg2.shape[2],
2)) # here we dont use any v4 data
for i in range(reg2.shape[0]):
datatest[i, :, :, 0] = reg2[i]
datatest[i, :, :, 1] = reg2[
i] # forming a paır of test samples all from v1.
print(predicted.shape)
xx = model.test(args, datatest)
predicted = xx.reshape(
-1, 1225
) # predicted[test_index,:] fill predicted with current test_index
ground = np.reshape(
mal2, (-1, 1225)
) # fill the previously defined predicted array with current test data
sess.close()
gc.collect()
print(predicted.shape, ground.shape)
corr, pvalue = pearsonr(predicted.ravel(), ground.ravel())
mse = mean_squared_error(predicted, ground)
pcr.append(corr)
pv.append(pvalue)
mser.append(mse)
predicteddata.append(predicted)
wholepredicted.append(predicted)
np.savetxt("./pred/pred" + str(test_index[0]) + ".txt",
predicted,
delimiter=',')
np.savetxt("./gt/gt" + str(test_index[0]) + ".txt",
ground,
delimiter=',')
print("pearson corr ", corr)
print("pvalue ", pvalue)
print("mean square error ", mse)
end = datetime.datetime.now()
print(start, 'thıs ıs the start tıme')
print(end, 'thıs ıs the end tıme')
print('total tıme ıs', start - end)
gc.collect()
tf.reset_default_graph()
with open("./savedrestc/pcr rhv4 to v3last.txt", "w") as f:
for s in pcr:
f.write(str(s) + "\n")
with open("./savedrestc/mser rhv4 to v3last.txt", "w") as f:
for p in mser:
f.write(str(p) + "\n")
with open("./savedrestc/pv rhv4 to v3last.txt", "w") as f:
for v in pv:
f.write(str(v) + "\n")
with open("./predwhole/predwhole.txt", "w") as f:
for w in predicteddata:
f.write(str(w) + "\n")
#pv_mean=pv/5
pcr_mean = sum(pcr) / len(pcr)
print("length of pv is ", len(pv))
print("length of pcr is ", len(pcr))
mse_mean = sum(mser) / len(mser)
pv_mean = sum(pv) / len(pv)
print(predicted.shape, ground.shape)
corr, pvalue = pearsonr(predicted.ravel(), ground.ravel())
mse = mean_squared_error(predicted, ground)
print("pearson corr ", corr)
print("mean square error ", mse)
print("Now mean results ")
print("pearson corr_mean ", pcr_mean)
print("mean square error_mean ", mse_mean)
print("pv_mean is ", pv_mean)
end = datetime.datetime.now()
print(start, 'thıs ıs the start tıme')
print(end, 'thıs ıs the end tıme')
print('total tıme ıs', start - end)
#np.savetxt("./predwhole/predicted3d.txt.txt",predicteddata)
predicteddata2d = predicteddata[1:2]
predicted2d = np.reshape(predicteddata2d, (30, 1225))
np.savetxt("./predwhole/2dpredicted.txt", predicted2d)
np.save("./predwhole/wholepredicteddata.npy", wholepredicted)
def main():
start = datetime.datetime.now()
if not os.path.exists(args.checkpoint_dir):
os.makedirs(args.checkpoint_dir)
if not os.path.exists(args.sample_dir):
os.makedirs(args.sample_dir)
if not os.path.exists(args.test_dir):
os.makedirs(args.test_dir)
v1=np.load('graph_view1.npy')#150, 35, 35
v4=np.load('graph_view2.npy')
n_splits = 5
predicted = np.zeros(shape=(int(v1.shape[0]/n_splits),v1.shape[1]*v1.shape[2]))# full predicted dataset
ground = np.zeros(shape=(int(v1.shape[0]/n_splits),v1.shape[1]*v1.shape[2])) # structure for test data
predicted2 = np.zeros(shape=(int(v1.shape[0]/n_splits),v1.shape[1]*v1.shape[2]))
ground2 = np.zeros(shape=(int(v1.shape[0]/n_splits),v1.shape[1]*v1.shape[2]))
from sklearn.model_selection import KFold
loo=KFold(n_splits=5)
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
session = tf.Session(config=config)
option=None
pcr=[]
mser=[]
pv=[]
predicteddata=[]
wholepredicted=[]
pcr2=[]
mser2=[]
pv2=[]
predicteddata2=[]
wholepredicted2=[]
for train_index, test_index in loo.split(v1):
tf.reset_default_graph()
with tf.Session() as sess:
model = graph2graph(sess, batch_size=args.batch_size,
checkpoint_dir=args.checkpoint_dir, sample_dir=args.sample_dir,test_dir=args.test_dir,train_dir=args.train_dir,graph_size=args.graph_size,output_size=args.output_size, dataset=args.dataset)
reg1, reg2 = v1[train_index], v1[test_index]# reg1 ıs v1 traın dataset, reg2 v1 testdataset
print(np.shape(reg1),'reg1')
print(np.shape(reg2),'reg2')
mal1, mal2 = v4[train_index], v4[test_index]#mal1:v4 traın dataset,mal2 v4 testdataset
datatrain=np.zeros((reg1.shape[0],reg1.shape[1],reg1.shape[2],2)) # forming a paır of traın samples
print('now the test index is', test_index)
for i in range(reg1.shape[0]):
datatrain[i,:,:,0]=reg1[i] ###reg1=train v1 reg ıs A and mal ıs B
datatrain[i,:,:,1]=mal1[i]
model.train(args, datatrain)
# prediction in the first direction
datatest=np.zeros((reg2.shape[0],reg2.shape[1],reg2.shape[2],2))
for i in range(reg2.shape[0]):
datatest[i,:,:,0]=reg2[i]
datatest[i,:,:,1]=reg2[i]# forming a paır of test samples, first direction prediction.
print(predicted.shape)
xx=model.test(args, datatest)
predicted = xx.reshape(-1,v1.shape[1]*v1.shape[2])
ground = np.reshape(mal2, (-1,v1.shape[1]*v1.shape[2]))
print(predicted.shape,ground.shape)
corr,pvalue = pearsonr(predicted.ravel(),ground.ravel())
mse=mean_squared_error(predicted,ground)
pcr.append(corr)
pv.append(pvalue)
mser.append(mse)
predicteddata.append(predicted)
wholepredicted.append(predicted)
print("pearson corr ", corr)
print("pvalue ", pvalue)
print("mean square error ",mse)
# prediction in opposite direction (second direction)
print('###### begin B to A#####')
datatest2=np.zeros((mal2.shape[0],mal2.shape[1],mal2.shape[2],2)) #
for i in range(reg2.shape[0]):
datatest2[i,:,:,0]=mal2[i]
datatest2[i,:,:,1]=mal2[i]# forming a paır of test samples.
print(predicted2.shape)
xx=model.test2(args, datatest2)
predicted2 = xx.reshape(-1,v1.shape[1]*v1.shape[2])
ground2 = np.reshape(reg2, (-1,v1.shape[1]*v1.shape[2]))
sess.close()
gc.collect()
print(predicted2.shape,ground2.shape)
corr2,pvalue2 = pearsonr(predicted2.ravel(),ground2.ravel())
mse2=mean_squared_error(predicted2,ground2)
pcr2.append(corr2)
pv2.append(pvalue2)
mser2.append(mse2)
predicteddata2.append(predicted2)
wholepredicted2.append(predicted2)
print("pearson corr ", corr2)
print("pvalue ", pvalue2)
print("mean square error ",mse2)
print('###### end of B to A#######')
end = datetime.datetime.now()
print (start, 'thıs ıs the start tıme')
print (end, 'thıs ıs the end tıme')
gc.collect()
tf.reset_default_graph()
pcr_mean=sum(pcr) / len(pcr)
print("length of pv is ", len(pv))
print("length of pcr is ", len(pcr))
mse_mean=sum(mser) / len(mser)
pv_mean =sum(pv) / len(pv)
print(predicted.shape,ground.shape)
print("****Now mean results A to B**** ")
print("pearson corr_mean ", pcr_mean)
print("mean square error_mean ",mse_mean)
print("pv_mean is ",pv_mean)
end = datetime.datetime.now()
print (start, 'thıs ıs the start tıme')
print (end, 'thıs ıs the end tıme')
np.save("wholepredicted_atob.npy", wholepredicted)
print('################## beginning of second part B to A #############################################')
pcr_mean2=sum(pcr2) / len(pcr2)
mse_mean2=sum(mser2) / len(mser2)
pv_mean2 =sum(pv2) / len(pv2)
print("****Now mean results for B to A **** ")
print("pearson corr_mean_B to A ", pcr_mean2)
print("mean square error_mean_B to A ",mse_mean2)
print("pv_mean is ",pv_mean2)
np.save("wholepredicted_btoa.npy", wholepredicted2)
print('################## end of second part B to A #############################################')
end = datetime.datetime.now()
print (start, 'thıs ıs the start tıme')
print (end, 'thıs ıs the end tıme')
print ('total tıme ıs', start-end)