[Python] Convolutional Neural Networks.ipynb

Kernel: Python 3

Convolutional Neural Networks

---

Data Preprocessing is done manually.

In our case, we:

• We created a dataset/ folder

• Then created training_set/ and test_set/ subfolders inside dataset/ folder

• Inside training_set/ and test_set/ subfolders we created cats/ and dogs/ subfolders

• Then we have placed 4000 cat pictures index 1-4000 in dataset/training_set/cats

• Then we have placed 4000 dog pictures index 1-4000 in dataset/training_set/dogs

• Then we have placed 1000 cat pictures index 4001-5000 in dataset/test_set/cats

• Then we have placed 1000 dog pictures index 4001-5000 in dataset/test_set/dogs

So now we have 4000 training examples for each class, and 1000 test examples for each class.

Our directory structure is something like this:

dataset/
    training_set/
        dogs/
            dog.1.jpg
            dog.2.jpg
            ...
        cats/
            cat.1.jpg
            cat.2.jpg
            ...
    test_set/
        dogs/
            dog.4001.jpg
            dog.4002.jpg
            ...
        cats/
            cat.4001.jpg
            cat.4002.jpg
            ...

If you want a bigger dataset then go here: [kaggle] Dogs vs. Cats(Create an algorithm to distinguish dogs from cats)

---

Additional Note: Fitting CNN to the image takes a lot of time and processing power. It would be better if you use tensorflow-gpu. Alternatively you can use any cloud service to train your network on. I personally used FloydHub. There are other alternative too such as AWS, Google Cloud, Paperspace etc. I will link few of them below.

• AWS

• Google Cloud

• Paperspace

• FloydHub

---

Importing the Keras libraries and packages

from keras.models import Sequential
from keras.layers import Conv2D
from keras.layers import MaxPooling2D
from keras.layers import Flatten
from keras.layers import Dense

Using TensorFlow backend.

Initialising the CNN

classifier = Sequential()

Step 1 - Convolution

classifier.add(Conv2D(32, (3, 3), input_shape=(64, 64, 3), activation="relu"))

Step 2 - Pooling

classifier.add(MaxPooling2D(pool_size = (2, 2)))

Adding a second convolutional layer followed by pooling

classifier.add(Conv2D(32, (3, 3), activation="relu"))
classifier.add(MaxPooling2D(pool_size = (2, 2)))

Step 3 - Flattening

classifier.add(Flatten())

Step 4 - Full connection

classifier.add(Dense(activation="relu", units=128))
classifier.add(Dense(activation="sigmoid", units=1))

Compiling the CNN

classifier.compile(optimizer = 'adam', loss = 'binary_crossentropy', metrics = ['accuracy'])

Part 2 - Fitting the CNN to the images

from keras.preprocessing.image import ImageDataGenerator

train_datagen = ImageDataGenerator(rescale = 1./255,
                                   shear_range = 0.2,
                                   zoom_range = 0.2,
                                   horizontal_flip = True)

test_datagen = ImageDataGenerator(rescale = 1./255)
training_set = train_datagen.flow_from_directory('/my_data/dataset/training_set',
                                                 target_size = (64, 64),
                                                 batch_size = 32,
                                                 class_mode = 'binary')

test_set = test_datagen.flow_from_directory('/my_data/dataset/test_set',
                                            target_size = (64, 64),
                                            batch_size = 32,
                                            class_mode = 'binary')
classifier.fit_generator(training_set,
                         epochs = 25,
                         validation_data = test_set,
                         validation_steps = 2000,
                         steps_per_epoch=250)

Using TensorFlow backend.

Found 8000 images belonging to 2 classes.
Found 2000 images belonging to 2 classes.
Epoch 1/25
250/250 [==============================] - 231s - loss: 0.6863 - acc: 0.5567 - val_loss: 0.6840 - val_acc: 0.5528
Epoch 2/25
250/250 [==============================] - 187s - loss: 0.6480 - acc: 0.6283 - val_loss: 0.6220 - val_acc: 0.6745
Epoch 3/25
250/250 [==============================] - 187s - loss: 0.5852 - acc: 0.6909 - val_loss: 0.5571 - val_acc: 0.7184
Epoch 4/25
250/250 [==============================] - 186s - loss: 0.5490 - acc: 0.7192 - val_loss: 0.5198 - val_acc: 0.7564
Epoch 5/25
250/250 [==============================] - 186s - loss: 0.5160 - acc: 0.7424 - val_loss: 0.5173 - val_acc: 0.7384
Epoch 6/25
250/250 [==============================] - 187s - loss: 0.4983 - acc: 0.7530 - val_loss: 0.4937 - val_acc: 0.7512
Epoch 7/25
250/250 [==============================] - 186s - loss: 0.4815 - acc: 0.7645 - val_loss: 0.5162 - val_acc: 0.7542
Epoch 8/25
250/250 [==============================] - 187s - loss: 0.4633 - acc: 0.7740 - val_loss: 0.4748 - val_acc: 0.7775
Epoch 9/25
250/250 [==============================] - 187s - loss: 0.4426 - acc: 0.7883 - val_loss: 0.4785 - val_acc: 0.7754
Epoch 10/25
250/250 [==============================] - 186s - loss: 0.4287 - acc: 0.7985 - val_loss: 0.4653 - val_acc: 0.7910
Epoch 11/25
250/250 [==============================] - 187s - loss: 0.4228 - acc: 0.7988 - val_loss: 0.4506 - val_acc: 0.7892
Epoch 12/25
250/250 [==============================] - 186s - loss: 0.4095 - acc: 0.8139 - val_loss: 0.4646 - val_acc: 0.7879
Epoch 13/25
250/250 [==============================] - 187s - loss: 0.3999 - acc: 0.8183 - val_loss: 0.4876 - val_acc: 0.7860
Epoch 14/25
250/250 [==============================] - 187s - loss: 0.3766 - acc: 0.8283 - val_loss: 0.4700 - val_acc: 0.7891
Epoch 15/25
250/250 [==============================] - 186s - loss: 0.3657 - acc: 0.8390 - val_loss: 0.4514 - val_acc: 0.8093
Epoch 16/25
250/250 [==============================] - 187s - loss: 0.3545 - acc: 0.8425 - val_loss: 0.4668 - val_acc: 0.7999
Epoch 17/25
250/250 [==============================] - 187s - loss: 0.3497 - acc: 0.8442 - val_loss: 0.4651 - val_acc: 0.7965
Epoch 18/25
250/250 [==============================] - 187s - loss: 0.3263 - acc: 0.8598 - val_loss: 0.4673 - val_acc: 0.8044
Epoch 19/25
250/250 [==============================] - 187s - loss: 0.3163 - acc: 0.8635 - val_loss: 0.4659 - val_acc: 0.8103
Epoch 20/25
250/250 [==============================] - 187s - loss: 0.2997 - acc: 0.8707 - val_loss: 0.4745 - val_acc: 0.8096
Epoch 21/25
250/250 [==============================] - 186s - loss: 0.2863 - acc: 0.8751 - val_loss: 0.5066 - val_acc: 0.7871
Epoch 22/25
250/250 [==============================] - 187s - loss: 0.2821 - acc: 0.8746 - val_loss: 0.4680 - val_acc: 0.8089
Epoch 23/25
250/250 [==============================] - 186s - loss: 0.2604 - acc: 0.8895 - val_loss: 0.4885 - val_acc: 0.8130
Epoch 24/25
250/250 [==============================] - 187s - loss: 0.2497 - acc: 0.8962 - val_loss: 0.4733 - val_acc: 0.8087
Epoch 25/25
250/250 [==============================] - 186s - loss: 0.2414 - acc: 0.8988 - val_loss: 0.5130 - val_acc: 0.8076

<keras.callbacks.History at 0x7fae3a700400>