problems if classes are more than six

[johnrobertus]

I am using: Multiclass_classification.ipynb
I have seen there is a output_classes = 4 which means you have four classes.

Now i want to use this Code for 10 classes. So I changed output_classes = 10. And I got this error:
ValueError: Only one class present in y_true. ROC AUC score is not defined in that case.

However, I tried this code with six classes (output_classes = 6) and it works fine. What should I change for more than six classes?

[MuhammedBuyukkinaci]

Thanks for using my repository.

I guess it is due to the fact that your batch size (steps = 8) is lower than 10. You should change steps=8 to 16 or 32 (maybe more). Also, can you give me more details on your class distribution, or could you provide countplot of your target for me?

[johnrobertus]

1. increasing the step size does not help unfortunately

2. class distribution:

import glob
#Train data
train = []
train_labels = []
files = glob.glob ("/Users/johnrobertus/Desktop/100DAGM/class5_non_anomaly_100/*.PNG") # your image path
for myFile in files:
image = cv2.imread (myFile)

image = cv2.resize(image, (IMG_SIZE_ALEXNET,IMG_SIZE_ALEXNET))
image = np.resize(image, (IMG_SIZE_ALEXNET, IMG_SIZE_ALEXNET, 3))
train.append (image)
train_labels.append([1, 0, 0, 0, 0, 0, 0, 0, 0, 0])

files = glob.glob ("/Users/johnrobertus/Desktop/100DAGM/class5_anomaly_100/*.png")
for myFile in files:
image = cv2.imread (myFile)

image = cv2.resize(image, (IMG_SIZE_ALEXNET,IMG_SIZE_ALEXNET))
image = np.resize(image, (IMG_SIZE_ALEXNET, IMG_SIZE_ALEXNET, 3))
train.append (image)
train_labels.append([0, 1, 0, 0, 0, 0, 0, 0, 0, 0])

files = glob.glob ("/Users/johnrobertus/Desktop/100DAGM/class4_non_anomaly/*.png")
for myFile in files:
image = cv2.imread (myFile)

image = cv2.resize(image, (IMG_SIZE_ALEXNET,IMG_SIZE_ALEXNET))
image = np.resize(image, (IMG_SIZE_ALEXNET, IMG_SIZE_ALEXNET, 3))
train.append (image)
train_labels.append([0, 0, 1, 0, 0, 0, 0, 0, 0, 0])

files = glob.glob ("/Users/johnrobertus/Desktop/100DAGM/class4_anomaly_100/*.png")
for myFile in files:
image = cv2.imread (myFile)

image = cv2.resize(image, (IMG_SIZE_ALEXNET,IMG_SIZE_ALEXNET))
image = np.resize(image, (IMG_SIZE_ALEXNET, IMG_SIZE_ALEXNET, 3))
train.append (image)
train_labels.append([0, 0, 0, 1, 0, 0, 0, 0, 0, 0])

files = glob.glob ("/Users/johnrobertus/Desktop/100DAGM/class3_non_anomaly_100/*.png")
for myFile in files:
image = cv2.imread (myFile)

image = cv2.resize(image, (IMG_SIZE_ALEXNET,IMG_SIZE_ALEXNET))
image = np.resize(image, (IMG_SIZE_ALEXNET, IMG_SIZE_ALEXNET, 3))
train.append (image)
train_labels.append([0, 0, 0, 0, 1, 0, 0, 0, 0, 0])

files = glob.glob ("/Users/johnrobertus/Desktop/100DAGM/class3_anomaly_100/*.png")
for myFile in files:
image = cv2.imread (myFile)

image = cv2.resize(image, (IMG_SIZE_ALEXNET,IMG_SIZE_ALEXNET))
image = np.resize(image, (IMG_SIZE_ALEXNET, IMG_SIZE_ALEXNET, 3))
train.append (image)
train_labels.append([0, 0, 0, 0, 0, 1, 0, 0, 0, 0])


files = glob.glob ("/Users/johnrobertus/Desktop/100DAGM/class2_non_anomaly_100/*.png")

for myFile in files:
image = cv2.imread (myFile)

image = cv2.resize(image, (IMG_SIZE_ALEXNET,IMG_SIZE_ALEXNET))
image = np.resize(image, (IMG_SIZE_ALEXNET, IMG_SIZE_ALEXNET, 3))
train.append (image)
train_labels.append([0, 0, 0, 0, 0, 0, 1, 0, 0, 0])

files = glob.glob ("/Users/johnrobertus/Desktop/100DAGM/class2_anomaly_100/*.png")

for myFile in files:
image = cv2.imread (myFile)

image = cv2.resize(image, (IMG_SIZE_ALEXNET,IMG_SIZE_ALEXNET))
image = np.resize(image, (IMG_SIZE_ALEXNET, IMG_SIZE_ALEXNET, 3))
train.append (image)
train_labels.append([0, 0, 0, 0, 0, 0, 0, 1, 0, 0])

files = glob.glob ("/Users/johnrobertus/Desktop/100DAGM/class1_non_anomaly_100/*.PNG")

for myFile in files:
image = cv2.imread (myFile)

image = cv2.resize(image, (IMG_SIZE_ALEXNET,IMG_SIZE_ALEXNET))
image = np.resize(image, (IMG_SIZE_ALEXNET, IMG_SIZE_ALEXNET, 3))
train.append (image)
train_labels.append([0, 0, 0, 0, 0, 0, 0, 0, 1, 0])

files = glob.glob ("/Users/johnrobertus/Desktop/100DAGM/class1_anomaly_100/*.png")

for myFile in files:
image = cv2.imread (myFile)

image = cv2.resize(image, (IMG_SIZE_ALEXNET,IMG_SIZE_ALEXNET))
image = np.resize(image, (IMG_SIZE_ALEXNET, IMG_SIZE_ALEXNET, 3))
train.append (image)
train_labels.append([0, 0, 0, 0, 0, 0, 0, 0, 0, 1])

train = np.array(train,dtype='float32') #as mnist
train_labels = np.array(train_labels,dtype='int32') #as mnist

[MuhammedBuyukkinaci]

If your class distribution is balanced(which I understood), you don't have to use ROC for evaluation. Accuracy and LogLoss are fine. I just used it to use it.

[johnrobertus]

thank you that worked for me

[MuhammedBuyukkinaci]

Great to hear it worked properly. Please like the repository if you like it.

[johnrobertus]

I have another unrelated question:

I ran the same code on a dataset of 50k images per class with a total amount of six classes in a high configuration cluster. After a couple of minutes I received a memory error:

Traceback (most recent call last):
File "/var/lib/condor/execute/dir_23348/condor_exec.exe", line 97, in
x, x_test, y, y_test = train_test_split(train,train_labels,test_size=0.2,train_size=0.8, random_state = 42)
File "/usr/local/lib/python3.6/dist-packages/sklearn/model_selection/_split.py", line 2212, in train_test_split
safe_indexing(a, test)) for a in arrays))
File "/usr/local/lib/python3.6/dist-packages/sklearn/model_selection/_split.py", line 2212, in
safe_indexing(a, test)) for a in arrays))
File "/usr/local/lib/python3.6/dist-packages/sklearn/utils/init.py", line 216, in safe_indexing
return X.take(indices, axis=0)
MemoryError

As a consequence I reduced the amount per class gradually and it only worked with an amount of under 500 images per class. I would've liked to run this code with more data. Is there something I have to change in the code for larger datasets?

[MuhammedBuyukkinaci]

As I understood, you became out of memory, which means your RAM isn't enough to load. You can reduce your data or you may use
import gc; gc.collect()
in your code to clean cache continuously.

[HostGuest]

please i wanna ask , i have image classification using cnn for 5 classes ( cloudy , snowy , foggy , sunny, rainy ) in each folder i have about 1300 images , what does it mean by image label ( is that mean : rainy1.jpg , rainy2.jpg ) the names of images or somthing else like ( the real category of image ( snowy ,foggy,rainy......... ) sorry i'm begginer

[iamsuzank]

i cant find the code to convert training and testing images ( you have provided the link of kaggle) into numpy array.please provide this code

[MuhammedBuyukkinaci]

Hi, I hope the code below is what you need.

#Labeling data
def label_img(img):
    word_label = img.split('_')[0]
    if word_label == 'glass': return [1,0]
    elif word_label == 'table': return [0,1]

#Function for importing data(images) from train directory.
def create_train_data():
    training_data = []
    for img in tqdm(os.listdir(TRAIN_DIR)):
        label = label_img(img)
        path = os.path.join(TRAIN_DIR,img)
        img = cv2.imread(path,1)
        img = cv2.resize(img, (IMG_SIZE,IMG_SIZE))
        training_data.append([np.array(img),np.array(label)])
    shuffle(training_data)
    np.save('train_data_bi.npy', training_data)
    return training_data

[naufalirfani]

How to make test_data_mc.npy?

[MuhammedBuyukkinaci]

Download data from kaggle. Put them in a directory. Use the function below to create test_data_mc.npy

#Function for importing data(images) from train directory.
def create_train_data():
    training_data = []
    for img in tqdm(os.listdir(TRAIN_DIR)):
        label = label_img(img)
        path = os.path.join(TRAIN_DIR,img)
        img = cv2.imread(path,1)
        img = cv2.resize(img, (IMG_SIZE,IMG_SIZE))
        training_data.append([np.array(img),np.array(label)])
    shuffle(training_data)
    np.save('train_data_bi.npy', training_data)
    return training_data

[naufalirfani]

I try to train with 42 classes so i change the step_size to 42, but i get an error "Tensorflow Serving | Input to reshape is a tensor with 10452 values, but the requested shape has 9216." What should i do?

[MuhammedBuyukkinaci]

If your target has 42 classes, output_classes variable should be 42 .

[naufalirfani]

what can i do to increase the accuracy? because i train with 42 classes, my accuracy is very low

[MuhammedBuyukkinaci]

You can augment your training data using ImageDataGenerator, which is under tensorflow.keras. You can use a pretrained neural network to initialize your training.