def getROC(input_file):
global nbP
global nbN
y_class = []
y_score = []
files = basicOperations.getFiles(input_file, "txt")
for inputFile in files:
f = open(inputFile, 'r')
lines = f.readlines()
for line in lines:
words = line.split()
y_class.append(int(words[-1]))
y_score.append(float(words[-2]))
if words[-1] == "0":
nbN += 1
elif words[-1] == "1":
nbP += 1
return metrics.roc_curve(y_class, y_score)
# 1,1,1,1,1,1,1,1]
idx = []
for i in range(len(feature_select)):
if feature_select[i]:
idx.append(i)
##########################################################################
##########################################################################
## 7/10 new
## I. Load all datasets
num_candi = []
if 1:
files = basicOperations.getFiles(path,'txt')
learn_set = []
learn_class = []
learn_orig = []
j=0
for maList in files:
print "Load:",j,maList
N_error = 0
f = open(maList,'r')
lines = f.readlines()
rows = len(lines)
L = 0
if rows>0:
def loadAllDataset(pickleFile, train_folders, N_featPF):
pklFiles = basicOperations.getFiles(input_path, "cpkl")
# pattern1 = ".+(?P<slide>slide[0-9]).+cpkl"
pattern1 = ".+pickles/(?P<slide>[A-Za-z0-9]+)/.+cpkl"
toto1 = re.compile(pattern1)
# pattern2 = ".+/(?P<filename>[A-Za-z0-9]+_[A-Za-z0-9]+_[0-9]+).+cpkl"
pattern2 = ".+/(?P<filename>[A-Za-z0-9]+_[A-Za-z0-9]+).+cpkl"
toto2 = re.compile(pattern2)
feat = {}
count = 1
Nobjects = [0,0,0] # number of mitos, notmitos and others
for folder in train_folders:
mito = []
notmito = []
notannot = []
others = []
for pkl in pklFiles:
m = re.search(folder, pkl)
if m == None:
continue
pkltemp = open(pkl, 'rb')
feats = cPickle.load(pkltemp)
for i in range(len(feats)):
if feats[i][-1]==1:
mito.append(feats[i][:-3])
elif feats[i][-1]==2:
notmito.append(feats[i][:-3])
if feats[i][-1]==3:
notannot.append(feats[i][:-3])
# centers.append(feats[i][-3:-1])
# slides.append(slide)
pkltemp.close()
# idx_shuffle = shuffleSeq(len(others))
# for i in range(N_featPF):
# notannot.append(others[idx_shuffle[i]])
# for x in mito:
# mitos.append(x)
# for x in notmito:
# notmitos.append(x)
######################################
### check is there invalide (nan or inf) element
mito = np.array(mito)
notmito = np.array(notmito)
notannot = np.array(notannot)
toto = [mito, notmito, notannot]
for i in range(3):
if np.isnan(toto[i]).any():
nanElem = np.isnan(toto[i])
nanElemRow = np.unique(np.where(nanElem == True)[0])
infElemRow = np.unique(np.where(toto[i] >= np.finfo(np.float32).max)[0])
nanElemRow = np.unique(np.hstack((nanElemRow, infElemRow)))
arBool = np.ones(len(toto[i]))
for x in nanElemRow:
arBool[x] = 0
arBool = arBool.astype(np.bool)
toto[i] = toto[i][arBool]
## centers = centers[arBool]
## slides = slides[arBool]
## filenames = filenames[arBool]
mito = toto[0]
notmito = toto[1]
notannot = toto[2]
for i in range(3):
for j in range(len(toto[i])):
Nobjects[i] += 1
feat[count] = {}
feat[count]['folder'] = folder
feat[count]['features'] = toto[i][j]
if i==0:
feat[count]['annotation'] = 'mito'
elif i==1:
feat[count]['annotation'] = 'notmito'
else:
feat[count]['annotation'] = 'others'
count += 1
# f = open(pickleFile,'r')
# feat = pickle.load(f)
# f.close()
## get feature_name ##
# featureName = feat[feat.keys()[0]]['features'].keys()
print Nobjects
featureName = 0
return feat, featureName
if __name__ == "__main__":
######
# parameter
n_job = 4
input_path = "/home/seawave/work/database/train_40"
csv_path = "/home/seawave/work/database/mitos_atypia"
output_path = "/home/seawave/work/output/test3"
train_folders = ["A03","A04","A05","A07","A10","A11","A12","A14","A15",\
"A17","A18", "H03","H04","H05","H07","H10","H11","H12","H14","H15",\
"H17","H18"]
# train_folders = ["A03","A04","A05"]
images = basicOperations.getFiles(input_path, "tiff")
if 0: ### check not analysed images
images_not_analysed = []
pattern1 = ".+train_40/(?P<slide>[A-Za-z0-9]+)/(?P<filename>[A-Za-z0-9_]+).tiff"
toto = re.compile(pattern1)
for image in images:
tata = toto.match(image)
filename = tata.groupdict()['filename']
imout = filename+".png"
outfile = os.path.join(output_path, imout)
if not os.path.isfile(outfile):
images_not_analysed.append(image)
if len(images_not_analysed) == 0:
print "finished"
## directories
path_result = "/Volumes/Xiwei_MacExt/output/test2"
path_imin = "/Volumes/Xiwei_MacExt/cecog_data/train_40"
path_csv = "/Volumes/Xiwei_MacExt/databases/mitos_atypia"
path_out_mitosis = "/Volumes/Xiwei_MacExt/output/test2_crops"
# path_out_not_mitosis = "/Volumes/Xiwei_MacExt/output/test1_crop"
# path_result = "/home/seawave/work/output/test2_div2"
# path_imin = "/home/seawave/work/database/train_40"
# path_csv = "/home/seawave/work/database/mitos_atypia"
# path_out_mitosis = "/home/seawave/work/output/test2_crops"
## select mitosis count GTs
GT_list_all = basicOperations.getFiles(path_csv,'csv')
orig_image_all = basicOperations.getFiles(path_imin,'tiff')
results = basicOperations.getFiles(path_result,'png')
mitosis_set = []
not_mitosis_set = []
L1 = len(path_csv)
for filetemp1 in GT_list_all:
filetemp2 = filetemp1[L1+1:]
m = re.search("mitosis", filetemp2)
if m!=None:
m2 = re.search("not", filetemp2)
if m2!=None:
not_mitosis_set.append(filetemp1)
else:
mitosis_set.append(filetemp1)
return tf.Variable(initial, name)
#Convolution and Pooling
def conv2d(x, W):
return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME')
def max_pool_2x2(x):
return tf.nn.max_pool(x, ksize=[1, 2, 2, 1],
strides=[1, 2, 2, 1], padding='SAME')
if __name__ == "__main__":
txt_files = basicOperations.getFiles(input_path, "txt")
ASF_files = basicOperations.getFiles(input_path, "png", "ASF")
MACandi_files = basicOperations.getFiles(input_path, "png", "MaCandi")
for testIdx in range(148):
size_pool = size
#### Get training data
array_trainning_set, array_trainning_class, test_image_set = getTrainingData()
###################################################################3
#### SETUP CNN
# tf.set_random_seed(1234)
sess = tf.InteractiveSession()
x = tf.placeholder(tf.float32, [None, size*size])
y_ = tf.placeholder(tf.float32, [None, 2])
import basicOperations
## A window centered at GT coordinate
winSize = 231
winSizeR = winSize/2
winSizeR2 = winSize/2+1
cut_orig = True
cut_result = True
## directories
path_result = "/home/seawave/work/output/test1b"
results = basicOperations.getFiles(path_result,'txt')
mitosis = []
for result in results:
textTemp = open(result,'r')
lines = textTemp.readlines()
if len(lines)==0:
continue
i = 0
for line in lines:
tempfeat = []
words = line.split(' ')
