def fromTheInternetTransfer(numGraphs):
cwd = os.getcwd()
#clean(cwd)
### CONTROL ###
# create encessary dirs
outputDir = "exp7_filtered_ctrl"
create_filtered_dirs(outputDir)
# All wild graphs
resave_wild_graphs_for_classification("train",
outputDir,
traintest='train')
resave_wild_graphs_for_classification("validation",
outputDir,
traintest='val')
Classifier
graph_classification(cwd, 0, outputDir)
### INTERVENTION ###
# create encessary dirs
outputDir = "exp7_filtered_test"
create_filtered_dirs(outputDir)
# generated graphs
poss_sNs = [1, 2, 3]
for i in range(numGraphs):
sN = choice(poss_sNs)
print(sN)
create_multiData(i,
sN,
'train',
'random',
'multi',
'multi',
'g',
outputDir=outputDir)
# wild graphs
resave_wild_graphs_for_classification("validation", outputDir)
# Classifier
graph_classification(cwd, 0, outputDir)
def eval_ve(numGraphs, seriesNumber, graphType, folder='chartOcr_images'):
hits = 0
aHits = 0
ctr = 0
gt_image_to_corr_set = {}
ve_image_to_corr_set = {}
for i in range(numGraphs):
if seriesNumber == "random":
seriesNum = choice(posSNs)
else:
seriesNum = seriesNumber
ctr += 1
name, x1s, x2s, tstr, label_to_corr_map, col_to_corr_map = create_multiData(
i,
seriesNum,
"train",
graphType,
"multi",
"solid",
'pt',
outputDir2=folder,
noHard=True)
iname = name + ".png"
print(iname)
gt_c = Counter(label_to_corr_map.values())
gt_image_to_corr_set[iname] = sorted(gt_c.elements())
test_string, test_leg_set = process_img(folder + "/" + iname,
use_text_not_color=False)
corrList = []
for elem in test_leg_set:
corrList.append(elem.split(": ")[1])
ve_c = Counter(corrList)
ve_image_to_corr_set[iname] = sorted(ve_c.elements())
print(gt_c)
print(ve_c)
intersection_c = gt_c & ve_c
union_c = gt_c | ve_c
aHits += sum(intersection_c.values()) / sum(union_c.values())
if (ve_c == gt_c):
hits += 1
with open(folder + '_info/ground_truth.txt', 'w') as gt_file:
gt_file.write(json.dumps(gt_image_to_corr_set))
with open(folder + '_info/results.txt', 'w') as result_file:
result_file.write(str(hits / ctr) + "\n" + str(aHits / ctr))
return hits / ctr
def exp3(numGraphs, wild=False):
# generate test graphs
poss_sNs = [1, 2, 3]
CURRks = []
CNNks = []
KNEEDLEks = []
sNs = []
for i in range(numGraphs):
sN = choice(poss_sNs)
print(sN)
sNs.append(sN)
name = create_multiData(i, sN, 'train', 'random', 'multi', 'multi',
'exp1')
img = cv2.imread(name)
#img = sat_thresh_filter(img,40)
print('finished preprocessing')
CURRks.append(elbowM(img) - 1)
Image.fromarray(img).save('./exp3_placeholder.png')
#CNNks.append(find_k('./exp3_placeholder.png')-1)
KNEEDLEks.append(elbowM(img, kneedleBasic=True) - 1)
print(sNs)
print(CURRks)
print(CNNks)
print(KNEEDLEks)
CURRscore = 0
CNNscore = 0
KNEEDLEscore = 0
for i in range(len(sNs)):
if CURRks[i] == sNs[i]:
CURRscore += 1
if CNNks[i] == sNs[i]:
CNNscore += 1
if KNEEDLEks[i] == sNs[i]:
KNEEDLEscore += 1
CURR = CURRscore / len(sNs)
CNN = CNNscore / len(sNs)
KNEEDLE = KNEEDLEscore / len(sNs)
print(CURR, CNN, KNEEDLE)
# plot results
plt.style.use('default')
X1 = ['Current Method', 'CNN', 'Basic Kneedle Algorithm']
X2 = [CURR, CNN, KNEEDLE]
plt.bar(X1, X2)
plt.ylabel("K Choice Accuracy")
plt.show()
def exp1(numGraphs):
# generate test graphs
poss_sNs = [1, 2, 3]
ks = []
pks = []
sNs = []
for i in range(numGraphs):
sN = choice(poss_sNs)
print(sN)
sNs.append(sN)
name = create_multiData(i, sN, 'train', 'random', 'multi', 'multi',
'exp1')
img = cv2.imread(name)
ks.append(elbowM(img) - 1)
img = sat_thresh_filter(img, 40)
print('finished preprocessing')
pks.append(elbowM(img) - 1)
print(sNs)
print(ks)
print(pks)
pscore = 0
score = 0
for i in range(len(sNs)):
if ks[i] == sNs[i]:
score += 1
if pks[i] == sNs[i]:
pscore += 1
print(score / len(sNs))
print(pscore / len(sNs))
nPP = score / len(sNs)
PP = pscore / len(sNs)
# plot results
plt.style.use('default')
X1 = ['No Preprocessing', 'Preprocessing']
X2 = [nPP, PP]
plt.bar(X1, X2)
plt.ylabel("K Choice Accuracy")
plt.show()
def exp10(numGraphs):
ground_truth = []
ground_truth.append(("filename", "x axis label", "y axis label",
"title string", "legend and correlation"))
posSNs = [1, 2, 3]
axis_hits = 0
series_hits = 0
axis_hits_i = 0
series_hits_i = 0
hits = 0
hits_i = 0
for i in range(0, numGraphs):
leg_display_str = ""
name, x1s, x2s, tstr, label_to_corr_map = create_multiData(
i, choice(posSNs), "train", "random", "multi", "solid", 'pt')
iname = name + ".png"
display_string = "images/" + iname + ", x axis: " + x1s + ", y axis: " + x2s + ", title: " + tstr
leg_set = set()
for key in label_to_corr_map:
leg_set.add(key + ": " + label_to_corr_map[key])
leg_display_str = leg_display_str + ", " + key + ": " + label_to_corr_map[
key]
ground_truth.append((iname, x1s, x2s, tstr, leg_display_str))
#display_string = display_string + leg_display_str
#test_string,test_leg_set = process_img("images/" + iname)
test_string, test_leg_set, test_leg_set_i = process_img("images/" +
iname,
algo='old')
print(display_string)
print(leg_set)
print(test_string)
print(test_leg_set)
#print(test_string_i)
print(test_leg_set_i)
edds = editfast(display_string, test_string)
if edds < 6:
axis_hits = axis_hits + 1
else:
print('axis fail --------------------------------')
# edds_i = editfast(display_string, test_string_i)
# if edds_i < 6:
# axis_hits_i = axis_hits_i + 1
# else:
# print('axis fail i --------------------------------')
# control
oldsh = series_hits
for elem1 in leg_set:
for elem2 in test_leg_set:
edsm = editfast(elem1, elem2)
#print('edsm =' + str(edsm))
if edsm < 3:
series_hits = series_hits + (1 / len(leg_set))
break
if oldsh == series_hits:
print('series fail --------------------------------')
if not bool(test_leg_set):
print('empty set for legend')
# intervetion
oldsh = series_hits_i
for elem1 in leg_set:
for elem2 in test_leg_set_i:
edsm = editfast(elem1, elem2)
#print('edsm =' + str(edsm))
if edsm < 3:
series_hits_i = series_hits_i + (1 / len(leg_set))
break
if oldsh == series_hits_i:
print('series fail i --------------------------------')
if not bool(test_leg_set_i):
print('empty set for legend i')
# control
hits = (axis_hits + series_hits) / 2
#print("axis score: " + str(axis_hits/numGraphs))
print("series score: " + str(series_hits / numGraphs))
score = (hits / numGraphs) * 100
#print("total score: " + str(score) + "%")
# intervention
#hits_i = (axis_hits_i + series_hits_i)/2
#print("axis score: " + str(axis_hits_i/numGraphs))
print("series score: " + str(series_hits_i / numGraphs))
#score_i = (hits_i/numGraphs)*100
#print("total score i: " + str(score_i) + "%")
# write csv files
with open('./exp10.csv', 'w') as f:
writer = csv.writer(f, delimiter=',')
writer.writerows(ground_truth)
f.close()
# outputDir = "exp10"
# create_filtered_dirs(outputDir)
# poss_sNs = [1,2,3]
# for i in range(numGraphs):
# sN = choice(poss_sNs)
# print(sN)
# create_multiData(i, sN, 'train', 'random', 'multi', 'multi', 'exp10', pstyle='multi', outputDir2=outputDir)
#exp7(100)
def syntheticVariedTransfer(numGraphs):
cwd = os.getcwd()
# Control
# create necessary dirs
#clean(os.getcwd())
outputDir = "exp8_filtered_ctrl"
create_filtered_dirs(outputDir)
poss_sNs = [1, 2, 3]
for i in range(numGraphs):
sN = choice(poss_sNs)
print(sN)
create_multiData(i,
sN,
'train',
'random',
'multi',
'multi',
'g',
pstyle='multi',
outputDir=outputDir)
for i in range(numGraphs):
sN = choice(poss_sNs)
print(sN)
create_multiData(i,
sN,
'validation',
'random',
'multi',
'multi',
'g',
pstyle='multi',
outputDir=outputDir)
# Classifier
graph_classification(cwd, 0, outputDir)
# intervention
# create necessary dirs
#clean(os.getcwd())
outputDir = "exp8_filtered_test"
create_filtered_dirs(outputDir)
for i in range(numGraphs):
sN = choice(poss_sNs)
print(sN)
create_multiData(i,
sN,
'train',
'random',
'multi',
'multi',
'g',
pstyle='default',
outputDir=outputDir)
print("finished part 1")
for i in range(numGraphs):
sN = choice(poss_sNs)
print(sN)
create_multiData(i,
sN,
'validation',
'random',
'multi',
'multi',
'g',
pstyle='multi',
outputDir=outputDir)
# Classifier
graph_classification(cwd, 0, outputDir)
def exp9(numGraphs):
# create control directory
path = "./exp9_ctrl"
try:
os.mkdir(path)
except OSError:
print(
"Warning: Creation of the directory %s failed, might already exist"
% path)
# generate dataset to test OCR
posSNs = [1, 2, 3]
#display_test_dic = {}
#leg_test_dic = {}
axis_hits_OD = 0
axis_hits_algo = 0
for i in range(numGraphs):
#leg_display_str = ""
name, x1s, x2s, tstr, label_to_corr_map = create_multiData(
i,
choice(posSNs),
"train",
"random",
"multi",
"solid",
'pt',
outputDir2='exp9_ctrl')
iname = name + ".png"
display_string = name + ", x axis: " + x1s + ", y axis: " + x2s + ", title: " + tstr #subsitute: "path + iname" for "name" to match original structure from pipeline_testing.py
imgPath = path + "/" + iname
jpgimg = Image.open(imgPath).convert('RGB')
newimgp = imgPath[:len(imgPath) - 3] + 'jpg' # convert png to jpg
jpgimg.save(newimgp)
ocr = OCR(imgPath,
assign_labels(show_inference(detection_model, newimgp)))
text_dict_OD = ocr.crop()
display_string_OD = name
text_dict_algo = {
'x axis': ocr.xAxisLab,
'y axis': ocr.yAxisLab,
'title': ocr.title
}
display_string_algo = name
#print(text_dict_algo)
for elem in text_dict_OD:
if elem != 'legend' and text_dict_OD[elem] is not None:
display_string_OD = display_string_OD + ", " + elem + ": " + ' '.join(
text_dict_OD[elem])
for elem in text_dict_algo:
if elem != 'legend' and text_dict_algo[elem] is not None:
display_string_algo = display_string_algo + ", " + elem + ": " + text_dict_algo[
elem]
print(display_string)
print(display_string_OD)
print(display_string_algo)
edds_OD = editfast(display_string, display_string_OD)
if edds_OD < 6:
axis_hits_OD = axis_hits_OD + 1
else:
print('axis fail OD --------------------------------')
edds_algo = editfast(display_string, display_string_algo)
if edds_algo < 6:
axis_hits_algo = axis_hits_algo + 1
else:
print('axis fail algo --------------------------------')
print("OD score: " + str(axis_hits_OD / numGraphs))
print("algo score: " + str(axis_hits_algo / numGraphs))
'''