def downloadImages(dataset):
print('Start reading features')
with open(dataset) as f:
allImgs = []
shapes = []
allResults = []
notProcessed = 0
totalImgs = 0
correctShape = 0
for row in csv.DictReader(f):
if (float(row["likeRatio"]) > 1.):
continue
print(totalImgs)
totalImgs += 1
try:
image = imageProcess.Image(row["imgUrl"], True)
imageShape = image.getImageShape()
shapes.append(imageShape)
# squaredImage = imageShape[0] == imageShape[1]
# isRgb = imageShape[2] == 3;
# if (not squaredImage) or (not isRgb):
# continue
# image_rescaled = rescale(image.skimageImage, RESIZE_FACTOR, anti_aliasing=False, multichannel=True)
image_rescaled = resize(image.skimageImage, (TARGET_X, TARGET_Y),anti_aliasing=False)
except Exception as e:
notProcessed += 1
print(e)
continue
allImgs.append(image_rescaled)
allResults.append(float(row["likeRatio"]))
print("not processed: " + str(notProcessed/totalImgs))
slashIndex = dataset.find("/")
slashIndex += 1
plt.figure()
plt.plot(shapes)
plt.title('image shape distribution')
plt.ylabel('width')
plt.xlabel('height')
plt.savefig(f"datasets/{dataset[slashIndex:-4]}_distribution.png")
np.save(f"allImgs_{dataset[slashIndex:-4]}.npy", allImgs)
np.save(f"allResults_{dataset[slashIndex:-4]}.npy", allResults)
return allImgs, allResults
def extractFeaturesFromDataset(filename):
net = imageProcess.runFaceDetectDNN()
print('Start reading features')
with open(filename) as f:
listFeatureVectorsWithResult = []
notProcessed = 0
for row in csv.DictReader(f):
featureVector = defaultdict(float)
imageNotProcessed = False
for key in row: # each row is a dict
if (key == "timestamp"):
hourOfDay = datetime.fromtimestamp(int(row[key])).hour
between2and6 = (hourOfDay >= 2 and hourOfDay < 6)
between6and10 = (hourOfDay >= 6 and hourOfDay < 10)
between10and14 = (hourOfDay >= 10 and hourOfDay < 14)
between14and18 = (hourOfDay >= 14 and hourOfDay < 18)
between18and22 = (hourOfDay >= 18 and hourOfDay < 22)
between22and2 = (hourOfDay >= 22) or (hourOfDay < 2)
featureVector['between2and6'] = between2and6
featureVector['between6and10'] = between6and10
# featureVector['between10and14'] = between10and14
featureVector['between14and18'] = between14and18
featureVector['between18and22'] = between18and22
featureVector['between22and2'] = between22and2
if (key == "caption"):
# featureVector["captionLength"] = (len(row[key]))
featureVector["capContainsFood"] = 1 if "food" in row[
key].lower() else 0
featureVector["capContainsFollow"] = 1 if "follow" in row[
key].lower() else 0
featureVector[
"capContainsAd"] = 1 if "ad" in row[key].lower() else 0
# if key == "hashtags":
# hashtags = ast.literal_eval(row[key])
# hashtags = [n.strip() for n in hashtags]
# featureVector["numHash"] = 1 if len(hashtags) == 0 else 1./len(hashtags)
if key == "imgUrl":
try:
image = imageProcess.Image(row[key], True)
except Exception as e:
print(e)
imageNotProcessed = True
break
# featureVector["colourfulness"] = imageProcess.extractSectorsFeature(image, 20, 20)
faceInfo = imageProcess.extractFaceInfo(image, net)
# featureVector["numFaces"] = imageProcess.extractNumFaces(faceInfo)
featureVector[
"percentageFaces"] = imageProcess.extractTotalPercentAreaFaces(
faceInfo)
if (key == "likeRatio" or key == "likeCount"
or key == "commentCount" or key == "timestamp"):
continue
# this should fail all the time we have a string as the value feature
# probably bad style but python has no better way to check if
# a string contains a float or not
try:
val = float(row[key])
featureVector[key] = val
except:
continue
if (imageNotProcessed):
notProcessed += 1
continue
label = float(row["likeRatio"])
listFeatureVectorsWithResult.append((featureVector, label))
print('Finished extracting features')
limitLen = 2 * int(len(listFeatureVectorsWithResult) / 3)
trainData = listFeatureVectorsWithResult[:limitLen]
testData = listFeatureVectorsWithResult[limitLen:]
print(
f"Not processed ratio: {notProcessed/len(listFeatureVectorsWithResult)}"
)
# plusOneCount = 0
# minusOneCount = 0
# for data in trainData:
# if data[1] == 1: plusOneCount+=1
# else: minusOneCount+=1
# for data in testData:
# if data[1] == 1: plusOneCount+=1
# else: minusOneCount+=1
# print(plusOneCount)
# print(minusOneCount)
# print(plusOneCount/(plusOneCount+minusOneCount))
return (trainData, testData)
def extractFeaturesFromDataset(filename):
print("PELE MEJOR QUE MARADONA!")
# net = imageProcess.runFaceDetectDNN()
print('Start reading features')
with open(filename) as f:
featureVectors = []
results = []
allImgs = []
allResults = []
shapes = []
notProcessed = 0
totalImgs = 0
correctShape = 0
a = True
for row in csv.DictReader(f):
if a:
print(row.keys())
a = False
print(totalImgs)
totalImgs += 1
featureVector = []
somethingFailed = False
for key in row: # each row is a dict
try:
if (key == "timestamp"):
hourOfDay = datetime.fromtimestamp(int(row[key])).hour
between2and6 = (hourOfDay >= 2 and hourOfDay < 6)
between6and10 = (hourOfDay >= 6 and hourOfDay < 10)
between10and14 = (hourOfDay >= 10 and hourOfDay < 14)
between14and18 = (hourOfDay >= 14 and hourOfDay < 18)
between18and22 = (hourOfDay >= 18 and hourOfDay < 22)
between22and2 = (hourOfDay >= 22) or (hourOfDay < 2)
featureVector.append(int(between2and6))
featureVector.append(int(between6and10))
featureVector.append(int(between10and14))
featureVector.append(int(between14and18))
featureVector.append(int(between18and22))
featureVector.append(int(between22and2))
dayOfWeek = ep_to_day(int(row[key]))
if dayOfWeek == "Sunday":
featureVector.append(1)
else:
featureVector.append(0)
if dayOfWeek == "Monday":
featureVector.append(1)
else:
featureVector.append(0)
if dayOfWeek == "Tuesday":
featureVector.append(1)
else:
featureVector.append(0)
if dayOfWeek == "Wednesday":
featureVector.append(1)
else:
featureVector.append(0)
if dayOfWeek == "Thursday":
featureVector.append(1)
else:
featureVector.append(0)
if dayOfWeek == "Friday":
featureVector.append(1)
else:
featureVector.append(0)
if dayOfWeek == "Saturday":
featureVector.append(1)
else:
featureVector.append(0)
elif (key == "accessibilityCaption"):
accessibilityCaption = row[key]
if "people" in accessibilityCaption:
featureVector.append(1)
else:
featureVector.append(0)
if "and" in accessibilityCaption:
featureVector.append(1)
else:
featureVector.append(0)
if "one" in accessibilityCaption:
featureVector.append(1)
else:
featureVector.append(0)
if "or" in accessibilityCaption:
featureVector.append(1)
else:
featureVector.append(0)
if "more" in accessibilityCaption:
featureVector.append(1)
else:
featureVector.append(0)
if "standing" in accessibilityCaption:
featureVector.append(1)
else:
featureVector.append(0)
if "nature" in accessibilityCaption:
featureVector.append(1)
else:
featureVector.append(0)
if "closeup" in accessibilityCaption:
featureVector.append(1)
else:
featureVector.append(0)
if "sitting" in accessibilityCaption:
featureVector.append(1)
else:
featureVector.append(0)
if "tree" in accessibilityCaption:
featureVector.append(1)
else:
featureVector.append(0)
if "photo" in accessibilityCaption:
featureVector.append(1)
else:
featureVector.append(0)
if "no" in accessibilityCaption:
featureVector.append(1)
else:
featureVector.append(0)
if "description" in accessibilityCaption:
featureVector.append(1)
else:
featureVector.append(0)
if "available" in accessibilityCaption:
featureVector.append(1)
else:
featureVector.append(0)
if "cloud" in accessibilityCaption:
featureVector.append(1)
else:
featureVector.append(0)
if "beard" in accessibilityCaption:
featureVector.append(1)
else:
featureVector.append(0)
if "mountain" in accessibilityCaption:
featureVector.append(1)
else:
featureVector.append(0)
if "child" in accessibilityCaption:
featureVector.append(1)
else:
featureVector.append(0)
if "playing" in accessibilityCaption:
featureVector.append(1)
else:
featureVector.append(0)
if "sports" in accessibilityCaption:
featureVector.append(1)
else:
featureVector.append(0)
if "sunglasses" in accessibilityCaption:
featureVector.append(1)
else:
featureVector.append(0)
if "on" in accessibilityCaption:
featureVector.append(1)
else:
featureVector.append(0)
if "grass" in accessibilityCaption:
featureVector.append(1)
else:
featureVector.append(0)
if "suit" in accessibilityCaption:
featureVector.append(1)
else:
featureVector.append(0)
if "selfie" in accessibilityCaption:
featureVector.append(1)
else:
featureVector.append(0)
if "crowd" in accessibilityCaption:
featureVector.append(1)
else:
featureVector.append(0)
if "1" in accessibilityCaption:
featureVector.append(1)
else:
featureVector.append(0)
if "person" in accessibilityCaption:
featureVector.append(1)
else:
featureVector.append(0)
if "wedding" in accessibilityCaption:
featureVector.append(1)
else:
featureVector.append(0)
if "baby" in accessibilityCaption:
featureVector.append(1)
else:
featureVector.append(0)
elif key == "imgUrl":
image = imageProcess.Image(row[key], True)
imageShape = image.getImageShape()
shapes.append((imageShape[0], imageShape[1]))
# squaredImage = imageShape[0] == imageShape[1]
# isRgb = imageShape[2] == 3;
# if (not squaredImage) or (not isRgb):
# continue
# image_rescaled = rescale(image.skimageImage, RESIZE_FACTOR, anti_aliasing=False, multichannel=True)
image_rescaled = resize(image.skimageImage, (TARGET_X, TARGET_Y),anti_aliasing=False)
# featureVector.append(imageProcess.extractSectorsFeature(image, 20, 20))
# faceInfo = imageProcess.extractFaceInfo(image, net)
# featureVector.append(imageProcess.extractNumFaces(faceInfo))
# featureVector.append(imageProcess.extractTotalPercentAreaFaces(faceInfo))
elif key == "likeRatio": # we will append the result at the end
continue #allResults.append(float(row[key]))
elif (key == "likeCount" or key == "commentCount"):
continue
featureVector.append(row[key])
elif (key == "isBusinessAcc"):
featureVector.append(int(row[key]=="True"))
elif (key == "isVerified"):
featureVector.append(int(row[key]=="True"))
elif (key == "hasChannel"):
featureVector.append(int(row[key]=="True"))
# this should fail all the time we have a string as the value feature
# probably bad style but python has no better way to check if
# a string contains a float or not
else:
try:
val = float(row[key])
featureVector[key] = val
except Exception as e:
continue
except Exception as e:
somethingFailed = True
notProcessed += 1
print(e)
break
if (somethingFailed):
continue
label = float(row["likeCount"])/float(row["userAverageLikes"])
allResults.append(label)
allImgs.append(image_rescaled)
featureVectors.append(featureVector)
slashIndex = filename.find("/")
slashIndex += 1
featureVectors = np.array(featureVectors)
allResults = np.array(allResults)
allImgs = np.array(allImgs)
plt.figure()
plt.title('image shape distribution')
plt.ylabel('width')
plt.xlabel('height')
plt.scatter(*zip(*shapes))
plt.savefig(f"datasets/{filename[slashIndex:-4]}_distribution.png")
np.save(f"allImgs_{filename[slashIndex:-4]}.npy", allImgs)
np.save(f"allResults_{filename[slashIndex:-4]}.npy", allResults)
np.save(f"featureVectors_{filename[slashIndex:-4]}.npy", featureVectors)
return allImgs, featureVectors, allResults
import imageprocess as imageProcess
import csv
from skimage.transform import rescale, resize, downscale_local_mean
RESIZE_FACTOR = 0.25
print('Start reading features')
with open('datasets/neuralnet-firstdataset.csv') as f:
allImgs = []
allResults = []
notProcessed = 0
totalImgs = 0
correctShape = 0
for row in csv.DictReader(f):
totalImgs += 1
try:
image = imageProcess.Image(row["imgUrl"], True)
if image.getImageShape() != (1080, 1080, 3):
continue
image_rescaled = rescale(image.skimageImage,
RESIZE_FACTOR,
anti_aliasing=False,
multichannel=True)
correctShape += 1
except Exception as e:
notProcessed += 1
continue
allImgs.append(image_rescaled)
allResults.append(float(row["likeRatio"]))
print(f"not processed: {notProcessed/totalImgs}")
print(f"correct shape total: {correctShape}")
print(f"correct shape ratio: {correctShape/totalImgs}")