import cv2

import pytesseract

import pandas as pd

from PIL import Image

import numpy as np

import shlex, subprocess

import shutil, os

from os import listdir

import sys,math

import PIL

import cv2

import numpy as np

from joblib import Parallel, delayed

def prepare_training_data(source_dir_list,target_dir):

counter = 0

for source in source_dir_list:

print(source)

file_names = listdir(source)

if counter == 0:

for file_name in file_names:

shutil.copy(source+file_name, target_dir+'Mis_'+file_name)

counter += 1

elif counter == 1:

for file_name in file_names:

shutil.copy(source+file_name, target_dir+'Nor_'+file_name)

counter += 1

elif counter == 2:

for file_name in file_names:

shutil.copy(source+file_name, target_dir+'Other_'+file_name)

def check_digit(input):

try:

float(input)

return True

except:

return False

def feature_extract_training(labeled_data, file):

y = -1

if file[:4] == 'Mis_':

y = 1

elif file[:4] == 'Nor_':

y = 0

else:

y = 2

try:

tb = pd.read_csv(labeled_data + file[:-4] + '-pred1-texts.csv')

except:

return

y_axis_value = tb[tb['type'] == 'y-axis-label']['text'].tolist()

y_axis_pos_y = tb[tb['type'] == 'y-axis-label']['y'].tolist()

y_axis_pos_y = [float(yy) for index, yy in enumerate(y_axis_pos_y) if check_digit(y_axis_value[index])]

y_axis_value = [float(yy) for yy in y_axis_value if check_digit(yy)]

y_axis_pairs = zip(y_axis_pos_y, y_axis_value)

res = sorted(y_axis_pairs, key=lambda x: x[0], reverse=True)

sorted(y_axis_pos_y)

steps = []

for index, yy in enumerate(y_axis_pos_y):

if index == len(y_axis_pos_y) - 1:

break

step = y_axis_pos_y[index + 1] - yy

steps.append(step)

if len(steps) > 1:

average_step = np.mean(steps)

else:

return

print(average_step)

increase_rates = []

for index, yy in enumerate(y_axis_value):

if index == len(y_axis_value) - 1:

break

increase_rate = (y_axis_value[index + 1] - yy) / average_step

increase_rates.append(increase_rate)

average_increase_rate = np.mean(increase_rates)

x_axis_pos_y = tb[tb['type'] == 'x-axis-label']['y'].tolist()

x_axis_pos_y = [float(xx) for xx in x_axis_pos_y]

average_x_axis_pos_y = np.mean(x_axis_pos_y)

if np.std(x_axis_pos_y) > 2:

# print("double x axes")

x_axis_pos_y = [float(xx) for xx in x_axis_pos_y if xx > average_x_axis_pos_y]

average_x_axis_pos_y = np.mean(x_axis_pos_y)

inference_flag_1 = 0

min_y_1 = 0

# print(res)

# print(average_step)

if len(res) == 0:

return []

if abs(res[0][0] - average_x_axis_pos_y) > abs(average_step / 2) and check_digit(res[1][1]) and res[-1][

0] < average_x_axis_pos_y:

# inference

# print('inference one')

inference_flag_1 = 1

# print(res[0][1])

# print(average_increase_rate)

# print(abs(res[0][0] - average_x_axis_pos_y))

min_y_1 = res[0][1] - average_increase_rate * (res[0][0] - average_x_axis_pos_y)

inference_flag_2 = 0

if not check_digit(res[0][1]) and check_digit(res[1][1]) and inference_flag_1 == 0:

# inference

# print('inference two')

inference_flag_2 = 1

min_y_1 = res[1][1] - average_increase_rate * (res[1][0] - x_axis_pos_y)

increase_rate_std = np.std(increase_rates)

inference_flag = 0

if inference_flag_1 == 0 and inference_flag_2 == 0:

min_y = res[0][1]

else:

inference_flag = 1

print('assign')

print(inference_flag_1)

print(min_y_1)

min_y = min_y_1

if math.isnan(min_y):

print(len(res))

min_y = 0

if np.isnan(increase_rate_std):

increase_rate_std = 0

if np.isnan(average_x_axis_pos_y):

return []

return [y, [inference_flag, (res[0][0] - average_x_axis_pos_y), min_y, increase_rate_std]]

def get_features_trainining(img_list,labeled_data):

X = []

y = []

files = []

for i in range(len(img_list)):

res = feature_extract_training(labeled_data,img_list[i])

if res != None and len(res) == 2 and len(res[1])==4:

X.append(res[1])

y.append(res[0])

files.append(img_list[i])

return X,y,files

def validation_prediction(labeled_data, file, y, annotated_graphs, target_dir):

outcome = ''

if file in listdir(annotated_graphs + 'Misleading\\') and y == 1:

outcome = 'TP'

shutil.copy(labeled_data + file, target_dir + 'TP\\' + file)

shutil.copy(labeled_data + file[:-4]+'-texts-all.csv', target_dir + 'TP\\' + file[:-4]+'-texts-all.csv')

elif file in listdir(annotated_graphs + 'Normal\\') or file in listdir(annotated_graphs + 'Other\\') and y == 1:

outcome = 'FP'

shutil.copy(labeled_data + file, target_dir + 'FP\\' + file)

shutil.copy(labeled_data + file[:-4]+'-texts-all.csv', target_dir + 'FP\\' + file[:-4]+'-texts-all.csv')

elif file in listdir(annotated_graphs + 'Normal\\') or file in listdir(annotated_graphs + 'Other\\') and y == 0:

outcome = 'TN'

shutil.copy(labeled_data + file, target_dir + 'TN\\' + file)

shutil.copy(labeled_data + file[:-4]+'-texts-all.csv', target_dir + 'TN\\' + file[:-4]+'-texts-all.csv')

elif file in listdir(annotated_graphs + 'Misleading\\') and y == 0:

outcome = 'FN'

shutil.copy(labeled_data + file, target_dir + 'FN\\' + file)

shutil.copy(labeled_data + file[:-4]+'-texts-all.csv', target_dir + 'FN\\' + file[:-4]+'-texts-all.csv')

def feature_extract_test(labeled_data, file):

try:

tb = pd.read_csv(labeled_data + file[:-4] + '-pred1-texts.csv')

except:

return

y_axis_value = tb[tb['type'] == 'y-axis-label']['text'].tolist()

y_axis_pos_y = tb[tb['type'] == 'y-axis-label']['y'].tolist()

y_axis_pos_y = [float(yy) for index, yy in enumerate(y_axis_pos_y) if check_digit(y_axis_value[index])]

y_axis_value = [float(yy) for yy in y_axis_value if check_digit(yy)]

y_axis_pairs = zip(y_axis_pos_y, y_axis_value)

res = sorted(y_axis_pairs, key=lambda x: x[0], reverse=True)

sorted(y_axis_pos_y)

steps = []

for index, yy in enumerate(y_axis_pos_y):

if index == len(y_axis_pos_y) - 1:

break

step = y_axis_pos_y[index + 1] - yy

steps.append(step)

if len(steps) > 1:

average_step = np.mean(steps)

else:

return

print(average_step)

increase_rates = []

for index, yy in enumerate(y_axis_value):

if index == len(y_axis_value) - 1:

break

increase_rate = (y_axis_value[index + 1] - yy) / average_step

increase_rates.append(increase_rate)

average_increase_rate = np.mean(increase_rates)

x_axis_pos_y = tb[tb['type'] == 'x-axis-label']['y'].tolist()

x_axis_pos_y = [float(xx) for xx in x_axis_pos_y]

average_x_axis_pos_y = np.mean(x_axis_pos_y)

if np.std(x_axis_pos_y) > 2:

x_axis_pos_y = [float(xx) for xx in x_axis_pos_y if xx > average_x_axis_pos_y]

average_x_axis_pos_y = np.mean(x_axis_pos_y)

inference_flag_1 = 0

min_y_1 = 0

return []

if abs(res[0][0] - average_x_axis_pos_y) > abs(average_step / 2) and check_digit(res[1][1]) and res[-1][

0] < average_x_axis_pos_y:

# inference

inference_flag_1 = 1

min_y_1 = res[0][1] - average_increase_rate * (res[0][0] - average_x_axis_pos_y)

inference_flag_2 = 0

if not check_digit(res[0][1]) and check_digit(res[1][1]) and inference_flag_1 == 0:

# inference

inference_flag_2 = 1

min_y_1 = res[1][1] - average_increase_rate * (res[1][0] - x_axis_pos_y)

increase_rate_std = np.std(increase_rates)

inference_flag = 0

if inference_flag_1 == 0 and inference_flag_2 == 0:

min_y = res[0][1]

else:

inference_flag = 1

print('assign')

print(inference_flag_1)

print(min_y_1)

min_y = min_y_1

if math.isnan(min_y):

print(len(res))

min_y = 0

if np.isnan(increase_rate_std):

increase_rate_std = 0

if np.isnan(average_x_axis_pos_y):

return []

return [inference_flag, (res[0][0] - average_x_axis_pos_y), min_y, increase_rate_std]

def get_features_test(img_list,labeled_data):

X = []

files = []

for i in range(len(img_list)):

res = feature_extract_training(labeled_data,img_list[i])

if res != None and len(res) == 2 and len(res[1])==4:

X.append(res)

files.append(img_list[i])

return X,files

def predict(random_forest,X_test,threshold,my_class):

predicted_proba = random_forest.predict_proba(X_test)

predicted = (predicted_proba[:, my_class] >= threshold).astype('int')

return predicted

def darknet_line_extrat(line):

terms = line.split(' ')

term_type = terms[0][:-1]

pos = [0] * 6

index = 0

for tt in terms:

if index == 0:

if tt.split(':')[0] == 'Enter' or len(tt.split(':')[0]) == 0 :

return -1

pos[0] = tt.split(':')[0]

index+=1

if tt == '':

continue

if '%' in tt:

pos[index] = int(tt.split('%')[0])

if 'left_x' in tt:

index = 2

try:

float(tt)

pos[index] = int(tt)

index += 1

except:

continue

return pos

#https://www.learnopencv.com/filling-holes-in-an-image-using-opencv-python-c/

def flood_fill_single(im_path):

# Read image

im_in = cv2.imread(im_path, cv2.IMREAD_GRAYSCALE);

# Threshold.

# Set values equal to or above 220 to 0.

# Set values below 220 to 255.

th, im_th = cv2.threshold(im_in, 220, 255, cv2.THRESH_BINARY_INV);

# Copy the thresholded image.

im_floodfill = im_th.copy()

# Mask used to flood filling.

# Notice the size needs to be 2 pixels than the image.

h, w = im_th.shape[:2]

mask = np.zeros((h + 2, w + 2), np.uint8)

# Floodfill from point (0, 0)

cv2.floodFill(im_floodfill, mask, (0, 0), 255);

# Invert floodfilled image

im_floodfill_inv = cv2.bitwise_not(im_floodfill)

# Combine the two images to get the foreground.

im_out = im_th | im_floodfill_inv

cv2.imwrite(im_path[:-4]+'_flood.png', im_out)

return

#https://stackoverflow.com/questions/35854197/how-to-use-opencvs-connected-components-with-stats-in-python

def check_fonts(im_path):

src = cv2.imread(im_path, cv2.IMREAD_GRAYSCALE)

# Threshold it so it becomes binary

ret, thresh = cv2.threshold(src, 127, 255, cv2.THRESH_BINARY)

# You need to choose 4 or 8 for connectivity type

connectivity = 4

# Perform the operation

output = cv2.connectedComponentsWithStats(thresh, connectivity, cv2.CV_32S)

font_list = []

# The second cell is the label matrix

labels = output[1]

# The third cell is the stat matrix

stats = output[2]

h_max = max([st[cv2.CC_STAT_HEIGHT] for st in stats])

w_max = max([st[cv2.CC_STAT_WIDTH] for st in stats])

for i, st in enumerate(stats):

h = st[cv2.CC_STAT_HEIGHT]

w = st[cv2.CC_STAT_WIDTH]

# remove dots in the figure

if h < h_max/15 and w < w_max/15:

continue

if h >=w:

font_list.append(h)

else:

font_list.append(w)

return sorted(font_list)[:-1]

def text_recognition(file, box_list):

lst = box_list

file_name = file.split('\\')[-1]

try:

os.mkdir('D:\\Misleading_Graph\\data\\tmp\\' + file_name + '\\')

except:

return -1

im = Image.open(open(file, "rb"))

l_list = []

font_list = []

for index, box in enumerate(lst[1]):

im1 = im.crop((lst[1][index] - 3, lst[2][index] - 3, lst[1][index] + lst[3][index] + 3,

lst[2][index] + lst[4][index] + 3))

im1.save('D:\\Misleading_Graph\\data\\tmp\\' + file_name + '\\tmp.png', 'PNG')

img = cv2.imread('D:\\Misleading_Graph\\data\\tmp\\' + file_name + '\\tmp.png')

if lst[7][index] == 'Number':

flood_fill_single('D:\\Misleading_Graph\\data\\tmp\\' + file_name + '\\tmp.png')

font_list = check_fonts('D:\\Misleading_Graph\\data\\tmp\\' + file_name + '\\tmp_flood.png')

df = pytesseract.image_to_data(img, lang='digitsall_layer', \

config='--psm 8 --oem 1 -c tessedit_char_whitelist=0123456789',

output_type='data.frame')

if len(font_list)>=3:

lst[8][index] = np.std(font_list)

else:

lst[8][index] = 0

else:

df = pytesseract.image_to_data(img, lang='eng', \

config='--psm 8 --oem 1 ', output_type='data.frame')

lst[8][index] = 0

string = df.sort_values(by=['conf'])['text'].iloc[-1]

conf = max(df['conf'].tolist())

lst[5][index] = string

lst[9][index] = conf

return lst

#https://stackoverflow.com/questions/35854197/how-to-use-opencvs-connected-components-with-stats-in-python

def letter_counter(im_path):

src = cv2.imread(im_path, cv2.IMREAD_GRAYSCALE)

# Threshold it so it becomes binary

ret, thresh = cv2.threshold(src, 0, 255, cv2.THRESH_BINARY)

# You need to choose 4 or 8 for connectivity type

connectivity = 4

# Perform the operation

output = cv2.connectedComponentsWithStats(thresh, connectivity, cv2.CV_32S)

font_list = []

# The second cell is the label matrix

labels = output[1]

# The third cell is the stat matrix

stats = output[2]

return len(stats)-1

def enlarge_corp(img_path,ratio):

img = Image.open(img_path)

width, height = img.size

img = img.resize((width*ratio, height*ratio), PIL.Image.ANTIALIAS)

img.save(img_path[:-4]+str(ratio)+'.png')

# thresholding

def thresholding(image):

return cv2.threshold(image, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)[1]

def ocr_preprocessing(img_path):

img = cv2.imread(img_path, cv2.IMREAD_GRAYSCALE)

thresh = thresholding(img)

cv2.imwrite(img_path[:-4]+'_prepro'+'.png', thresh)

def text_recognition_validate(file, box_list):

lst = box_list

file_name = file.split('/')[-1]

tmp_path = '/home/thuang12/Desktop/Misleading_Graph/data/tmp/'

try:

#os.mkdir('tmp/' + file_name + '/')

os.mkdir(tmp_path + file_name + '/')

except:

return -1

im = Image.open(open(file, "rb"))

l_list = []

font_list = []

for index, box in enumerate(lst[1]):

im1 = im.crop((lst[1][index] - 3, lst[2][index] - 3, lst[1][index] + lst[3][index] + 3,

lst[2][index] + lst[4][index] + 3))

#im1.save('tmp/' + file_name + '/tmp.png', 'PNG')

im1.save(tmp_path + file_name + '/tmp.png', 'PNG')

#img = cv2.imread('tmp/' + file_name + '/tmp.png')

img = cv2.imread(tmp_path + file_name + '/tmp.png')

#ocr_preprocessing('D:\\Misleading_Graph\\data\\tmp\\' + file_name + '\\tmp.png')

#flood_fill_single('tmp/' + file_name + '/tmp.png')

flood_fill_single(tmp_path + file_name + '/tmp.png')

#font_list = check_fonts('tmp/' + file_name + '/tmp_flood.png')

font_list = check_fonts(tmp_path + file_name + '/tmp_flood.png')

#letter_count = letter_counter('tmp/' + file_name + '/tmp_flood.png')

letter_count = letter_counter(tmp_path + file_name + '/tmp_flood.png')

if len(font_list)>=3:

#print(font_list)

lst[8][index] = np.std(font_list)

else:

lst[8][index] = 0

#df1 = pytesseract.image_to_data(img, lang='digitsall_layer', \

# config='--psm 8 --oem 1 -c tessedit_char_whitelist=0123456789',

# output_type='data.frame')

#pil_img = Image.open('tmp/' + file_name + '/tmp.png')

pil_img = Image.open(tmp_path + file_name + '/tmp.png')

rotated_pil_img = pil_img.rotate(270,expand=True)

#rotated_pil_img.save('tmp/' + file_name + '/tmp2.png')

rotated_pil_img.save(tmp_path + file_name + '/tmp2.png')

pil_img.close()

#img2 = cv2.imread('tmp/' + file_name + '/tmp2.png')

img2 = cv2.imread(tmp_path + file_name + '/tmp2.png')

df1 = pytesseract.image_to_data(img, lang='engrestrict_best', \

config='--psm 8 --oem 1 ', output_type='data.frame')

#df2 = pytesseract.image_to_data(img, lang='digits_comma', \

# config='--psm 8 --oem 1 ', output_type='data.frame')

df2 = pytesseract.image_to_data(img2, lang='engrestrict_best', \

config='--psm 8 --oem 1 ', output_type='data.frame')

#img2 = cv2.imread('D:\\Misleading_Graph\\data\\tmp\\' + file_name + '\\tmp_prepro.png')

#df3 = pytesseract.image_to_data(img2, lang='digitsall_layer', \

# config='--psm 8 --oem 1 -c tessedit_char_whitelist=0123456789',

# output_type='data.frame')

#df4 = pytesseract.image_to_data(img2, lang='eng1', \

# config='--psm 8 --oem 1 ', output_type='data.frame')

df = pd.concat([df1,df2])

df = df.sort_values(by=['conf'],ascending=False)

my_str = ''

conf = 0

#my_str = df['text'].iloc[0]

#conf = df['conf'].iloc[0]

distance = 999

print('here')

print(letter_count)

for ind, row in df.iterrows():

if row['conf'] == -1:

continue

t_str = str(row['text'])

if t_str[-1] == '.' or t_str[-1] == '|':

t_str = t_str[:-1]

if t_str.split('.')[-1] == '0':

t_str = str(int(float(t_str)))

if t_str == 'o':

t_str = str(0)

print(t_str)

if abs(len(str(t_str))-letter_count) <distance and len(t_str)>len(my_str):

my_str = t_str

conf = row['conf']

distance = abs(len(str(t_str))-letter_count)

print(my_str)

#print(letter_count)

lst[5][index] = my_str

lst[9][index] = conf

#print(lst[5])

return lst

def parse_line_text(text_list, target_dir):

file = text_list[0]

file_name = file.split('/')[-1]

lst = text_list[1]

lst = text_recognition(file, lst)

print(lst)

if lst == -1:

return

df = pd.DataFrame(zip(lst[0], lst[1], lst[2], lst[3], lst[4], lst[5], lst[6]),

columns=['id', 'x', 'y', 'width', 'height', 'text', 'type'])

df.to_csv(target_dir + file_name[:-4] + '-texts.csv', index=False)

lst2 = text_list[2]

df = pd.DataFrame(zip(lst[0], lst2[1],lst[9],lst[8]), columns=['id', 'prob', 'conf', 'font'])

df.to_csv(target_dir + file_name[:-4] + '-texts-prob.csv', index=False)

shutil.copy(file, target_dir + file_name)

print('done')

def parse_line_text_validate(text_list,org_img_dir, target_dir):

file = text_list[0]

file_name = file.split('/')[-1]

file = org_img_dir+file_name

print(file)

lst = text_list[1]

lst = text_recognition_validate(file, lst)

print(lst)

if lst == -1:

print('return')

return

df = pd.DataFrame(zip(lst[0], lst[1], lst[2], lst[3], lst[4], lst[5], lst[6]),

columns=['id', 'x', 'y', 'width', 'height', 'text', 'type'])

df.to_csv(target_dir + file_name[:-4] + '-texts.csv', index=False)

lst2 = text_list[2]

df = pd.DataFrame(zip(lst[0], lst2[1],lst[9],lst[8]), columns=['id', 'prob', 'conf', 'font'])

df.to_csv(target_dir + file_name[:-4] + '-texts-prob.csv', index=False)

shutil.copy(file, target_dir + file_name)

print('done')

def parse_lines_text_generated(annotation_dir,content):

counter = 0

text_list = []

for line in content:

text_id = []

text_type = []

text_prob = []

text_x = []

text_y = []

text_width = []

text_height = []

text_content = []

text_class = []

text_font = []

text_conf = []

try:

table = pd.read_csv(annotation_dir+line[:-4]+'_labels.csv')

text_list.append([line])

except:

continue

for index,row in table.iterrows():

text_id.append(index)

text_prob.append(1)

text_x.append(row['x'])

text_y.append(row['y'])

text_width.append(row['width'])

text_height.append(row['height'])

text_content.append(row['text'])

text_type.append('')

text_class.append('')

text_font.append(0)

text_conf.append(1)

lst = [text_id, text_x, text_y, text_width, text_height, text_content, text_type, text_class, text_font, text_conf]

text_list[-1].append(lst)

lst = [text_id, text_prob]

text_list[-1].append(lst)

print(counter)

return text_list

def parse_lines_text(content):

flag = 0

file = ''

counter = 0

text_list = []

text_id = []

text_type = []

text_prob = []

text_x = []

text_y = []

text_width = []

text_height = []

text_content = []

text_class = []

text_font = []

text_conf = []

file_name_list = []

for line in content:

if 'Predicted in' in line:

flag = 1

t_id = 0

if counter == 0:

file = line.split(': ')[0]

file_name = file.split('\\')[-1]

if counter > 0:

text_list.append([file])

lst = [text_id, text_x, text_y, text_width, text_height, text_content, text_type, text_class, text_font, text_conf]

text_list[-1].append(lst)

lst = [text_id, text_prob]

text_list[-1].append(lst)

text_id = []

text_type = []

text_prob = []

text_x = []

text_y = []

text_width = []

text_height = []

text_content = []

text_class = []

text_font = []

text_conf = []

counter += 1

file = line.split(': ')[0]

file_name = file.split('\\')[-1]

continue

if flag == 1:

im = Image.open(open(file, "rb"))

info = darknet_line_extrat(line[:-2])

if info == -1:

continue

text_id.append(t_id)

text_prob.append(info[1])

text_x.append(info[2])

text_y.append(info[3])

text_width.append(info[4])

text_height.append(info[5])

text_content.append('')

text_type.append('')

text_class.append(info[0])

text_font.append(0)

text_conf.append(0)

t_id += 1

print(counter)

return text_list

def subplot_crop(file, box_list, target_dir):

lst = box_list

file_name = file.split('/')[-1]

#print(file)

try:

im = Image.open(open(file, "rb"))

except:

return

l_list = []

for index, box in enumerate(lst[1]):

im1 = im.crop((lst[1][index] - 15, lst[2][index] - 15, lst[1][index] + lst[3][index] + 15,

lst[2][index] + lst[4][index] + 15))

#im1.save(target_dir + file_name[:-4] + '_'+str(index)+'.png', 'PNG')

im1.convert('RGB').save(target_dir + file_name[:-4] + '_'+str(index)+'.png', "PNG", optimize=True)

return

def parse_line_subplot(plot_list, target_dir):

file = plot_list[0]

#print(file)

file_name = file.split('/')[-1]

lst = plot_list[1]

lst = subplot_crop(file, lst,target_dir)

def parse_lines_subplot(content):

flag = 0

file = ''

counter = 0

subplot_list = []

subplot_id = []

subplot_type = []

subplot_prob = []

subplot_x = []

subplot_y = []

subplot_width = []

subplot_height = []

subplot_class = []

file_name_list = []

for line in content:

if 'Predicted in' in line:

flag = 1

t_id = 0

if counter == 0:

file = line.split(': ')[0]

file_name = file.split('\\')[-1]

if counter > 0:

subplot_list.append([file])

lst = [subplot_id, subplot_x, subplot_y, subplot_width, subplot_height, subplot_class]

subplot_list[-1].append(lst)

lst = [subplot_id, subplot_prob]

subplot_list[-1].append(lst)

subplot_id = []

subplot_prob = []

subplot_x = []

subplot_y = []

subplot_width = []

subplot_height = []

subplot_class = []

counter += 1

file = line.split(': ')[0]

file_name = file.split('\\')[-1]

continue

if flag == 1:

try:

im = Image.open(open(file, "rb"))

except:

continue

info = darknet_line_extrat(line[:-2])

if info == -1:

continue

subplot_id.append(t_id)

subplot_prob.append(info[1])

subplot_x.append(info[2])

subplot_y.append(info[3])

subplot_width.append(info[4])

subplot_height.append(info[5])

subplot_class.append(info[0])

t_id += 1

print(counter)

return subplot_list

def get_role(file_name,data_path,labeled_data):

if file_name[-4:] != '.png':

return

t2 = pd.read_csv(labeled_data + file_name[:-4] + '-texts-prob.csv')

if t2.shape[0] <2:

return

#command_line = 'powershell.exe wsl python /home/hzhuang/rev/run_text_role_classifier.py single ' + data_path + file_name

command_line = 'python /home/thuang12/Desktop/Misleading_Graph/rev/scripts/run_text_role_classifier.py single '+ data_path + file_name

args = command_line.split(' ')

p = subprocess.Popen(args,stdout=subprocess.PIPE,stderr=subprocess.PIPE)

print(p.communicate())

t1 = pd.read_csv(labeled_data+file_name[:-4]+'-pred1-texts.csv')

#t2 = pd.read_csv(labeled_data+file_name[:-4]+'-texts-prob.csv')

t_all = t1.merge(t2,on='id')

t_all.to_csv(labeled_data+file_name[:-4]+'-texts-all.csv')

def get_role_v(file_name,data_path,labeled_data):

if file_name[-4:] != '.png' or file_name[-10:] == '_check.png':

return

#t2 = pd.read_csv(labeled_data + file_name[:-4] + '-texts-prob.csv')

#if t2.shape[0] <2:

# return

command_line = 'powershell.exe wsl python /home/hzhuang/rev/run_text_role_classifier.py single ' + data_path + file_name

args = command_line.split(' ')

p = subprocess.Popen(args,stdout=subprocess.PIPE,stderr=subprocess.PIPE)

print(p.communicate())

#t1 = pd.read_csv(labeled_data+file_name[:-4]+'-pred1-texts.csv')

#t2 = pd.read_csv(labeled_data+file_name[:-4]+'-texts-prob.csv')

#t_all = t1.merge(t2,on='id')

#t_all.to_csv(labeled_data+file_name[:-4]+'-texts-all.csv')

def get_prob_dic(probability_dir):

#Compound_Prob

f1 = open(probability_dir + 'Comp.txt', "r")

lines_1 = f1.readlines()

f1.close()

f2 = open(probability_dir + 'Subplot.txt', "r")

lines_2 = f2.readlines

f2.close()

f3 = open(probability_dir + '3D.txt', "w")

lines_3 = f3.readlines

f3.close()

comp_dic = {}

for line in lines_1:

comp_dic[line.split(',')[0]] = line.split(',')[1]

subp_dic = {}

for line in lines_2:

subp_dic[line.split(',')[0]] = line.split(',')[1]

d3_dic = {}

for line in lines_3:

d3_dic[line.split(',')[0]] = line.split(',')[1]

return comp_dic, subp_dic, d3_dic

def get_prob_dic(file, comp_dic, subp_dic, d3_dic):

comp_prob = -1

subp_prob = -1

d3_prob = -1

if '_' in file:

comp_prob = comp_dic[file.split('_')[0] + '.png']

subp_prob = subp_dic[file]

else:

comp_prob = com_dic[file]

subp_prob = -1

d3_prob = d3_dic[file]

return comp_prob, subp_prob, d3_prob