TF2 0 CIFAR Improved
================ by Jawad Haider
# Install TensorFlow
# !pip install -q tensorflow-gpu==2.0.0-beta1
try:
%tensorflow_version 2.x # Colab only.
except Exception:
pass
import tensorflow as tf
print(tf.__version__)
`%tensorflow_version` only switches the major version: 1.x or 2.x.
You set: `2.x # Colab only.`. This will be interpreted as: `2.x`.
TensorFlow 2.x selected.
2.2.0-rc2
# additional imports
import numpy as np
import matplotlib.pyplot as plt
from tensorflow.keras.layers import Input, Conv2D, Dense, Flatten, Dropout, GlobalMaxPooling2D, MaxPooling2D, BatchNormalization
from tensorflow.keras.models import Model
# Load in the data
cifar10 = tf.keras.datasets.cifar10
(x_train, y_train), (x_test, y_test) = cifar10.load_data()
x_train, x_test = x_train / 255.0, x_test / 255.0
y_train, y_test = y_train.flatten(), y_test.flatten()
print("x_train.shape:", x_train.shape)
print("y_train.shape", y_train.shape)
Downloading data from https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz
170500096/170498071 [==============================] - 13s 0us/step
x_train.shape: (50000, 32, 32, 3)
y_train.shape (50000,)
number of classes: 10
# Build the model using the functional API
i = Input(shape=x_train[0].shape)
# x = Conv2D(32, (3, 3), strides=2, activation='relu')(i)
# x = Conv2D(64, (3, 3), strides=2, activation='relu')(x)
# x = Conv2D(128, (3, 3), strides=2, activation='relu')(x)
x = Conv2D(32, (3, 3), activation='relu', padding='same')(i)
x = BatchNormalization()(x)
x = Conv2D(32, (3, 3), activation='relu', padding='same')(x)
x = BatchNormalization()(x)
x = MaxPooling2D((2, 2))(x)
# x = Dropout(0.2)(x)
x = Conv2D(64, (3, 3), activation='relu', padding='same')(x)
x = BatchNormalization()(x)
x = Conv2D(64, (3, 3), activation='relu', padding='same')(x)
x = BatchNormalization()(x)
x = MaxPooling2D((2, 2))(x)
# x = Dropout(0.2)(x)
x = Conv2D(128, (3, 3), activation='relu', padding='same')(x)
x = BatchNormalization()(x)
x = Conv2D(128, (3, 3), activation='relu', padding='same')(x)
x = BatchNormalization()(x)
x = MaxPooling2D((2, 2))(x)
# x = Dropout(0.2)(x)
# x = GlobalMaxPooling2D()(x)
x = Flatten()(x)
x = Dropout(0.2)(x)
x = Dense(1024, activation='relu')(x)
x = Dropout(0.2)(x)
x = Dense(K, activation='softmax')(x)
model = Model(i, x)
# Compile
# Note: make sure you are using the GPU for this!
model.compile(optimizer='adam',
loss='sparse_categorical_crossentropy',
metrics=['accuracy'])
Epoch 1/50
1465/1563 [===========================>..] - ETA: 0s - loss: 1.2974 - accuracy: 0.5487
KeyboardInterrupt: ignored
# Fit with data augmentation
# Note: if you run this AFTER calling the previous model.fit(), it will CONTINUE training where it left off
batch_size = 32
data_generator = tf.keras.preprocessing.image.ImageDataGenerator(width_shift_range=0.1, height_shift_range=0.1, horizontal_flip=True)
train_generator = data_generator.flow(x_train, y_train, batch_size)
steps_per_epoch = x_train.shape[0] // batch_size
r = model.fit(train_generator, validation_data=(x_test, y_test), steps_per_epoch=steps_per_epoch, epochs=50)
Epoch 1/50
1562/1562 [==============================] - 27s 17ms/step - loss: 0.9854 - accuracy: 0.6597 - val_loss: 0.9380 - val_accuracy: 0.6898
Epoch 2/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.8444 - accuracy: 0.7101 - val_loss: 0.8461 - val_accuracy: 0.7158
Epoch 3/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.7506 - accuracy: 0.7444 - val_loss: 0.7562 - val_accuracy: 0.7485
Epoch 4/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.6825 - accuracy: 0.7695 - val_loss: 0.6062 - val_accuracy: 0.7959
Epoch 5/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.6324 - accuracy: 0.7856 - val_loss: 0.6161 - val_accuracy: 0.7971
Epoch 6/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.5944 - accuracy: 0.7997 - val_loss: 0.6473 - val_accuracy: 0.7832
Epoch 7/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.5502 - accuracy: 0.8109 - val_loss: 0.6116 - val_accuracy: 0.8007
Epoch 8/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.5210 - accuracy: 0.8227 - val_loss: 0.6694 - val_accuracy: 0.7861
Epoch 9/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.4947 - accuracy: 0.8300 - val_loss: 0.4850 - val_accuracy: 0.8358
Epoch 10/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.4690 - accuracy: 0.8407 - val_loss: 0.5492 - val_accuracy: 0.8174
Epoch 11/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.4480 - accuracy: 0.8479 - val_loss: 0.5357 - val_accuracy: 0.8212
Epoch 12/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.4283 - accuracy: 0.8523 - val_loss: 0.5085 - val_accuracy: 0.8319
Epoch 13/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.4125 - accuracy: 0.8577 - val_loss: 0.5201 - val_accuracy: 0.8308
Epoch 14/50
1562/1562 [==============================] - 26s 16ms/step - loss: 0.3929 - accuracy: 0.8662 - val_loss: 0.4446 - val_accuracy: 0.8510
Epoch 15/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.3779 - accuracy: 0.8694 - val_loss: 0.4738 - val_accuracy: 0.8506
Epoch 16/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.3706 - accuracy: 0.8721 - val_loss: 0.4617 - val_accuracy: 0.8504
Epoch 17/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.3470 - accuracy: 0.8812 - val_loss: 0.4172 - val_accuracy: 0.8627
Epoch 18/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.3405 - accuracy: 0.8818 - val_loss: 0.4572 - val_accuracy: 0.8587
Epoch 19/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.3312 - accuracy: 0.8868 - val_loss: 0.4150 - val_accuracy: 0.8654
Epoch 20/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.3203 - accuracy: 0.8880 - val_loss: 0.5443 - val_accuracy: 0.8273
Epoch 21/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.3108 - accuracy: 0.8919 - val_loss: 0.4421 - val_accuracy: 0.8605
Epoch 22/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.3017 - accuracy: 0.8964 - val_loss: 0.4778 - val_accuracy: 0.8537
Epoch 23/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.2978 - accuracy: 0.8975 - val_loss: 0.4370 - val_accuracy: 0.8621
Epoch 24/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.2830 - accuracy: 0.9023 - val_loss: 0.4270 - val_accuracy: 0.8676
Epoch 25/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.2796 - accuracy: 0.9035 - val_loss: 0.4009 - val_accuracy: 0.8748
Epoch 26/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.2714 - accuracy: 0.9069 - val_loss: 0.4017 - val_accuracy: 0.8719
Epoch 27/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.2663 - accuracy: 0.9080 - val_loss: 0.4199 - val_accuracy: 0.8669
Epoch 28/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.2547 - accuracy: 0.9121 - val_loss: 0.4094 - val_accuracy: 0.8703
Epoch 29/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.2574 - accuracy: 0.9110 - val_loss: 0.4227 - val_accuracy: 0.8698
Epoch 30/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.2464 - accuracy: 0.9141 - val_loss: 0.4117 - val_accuracy: 0.8649
Epoch 31/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.2439 - accuracy: 0.9157 - val_loss: 0.4096 - val_accuracy: 0.8758
Epoch 32/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.2411 - accuracy: 0.9159 - val_loss: 0.4118 - val_accuracy: 0.8705
Epoch 33/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.2330 - accuracy: 0.9194 - val_loss: 0.3841 - val_accuracy: 0.8764
Epoch 34/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.2312 - accuracy: 0.9201 - val_loss: 0.4127 - val_accuracy: 0.8708
Epoch 35/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.2272 - accuracy: 0.9209 - val_loss: 0.4259 - val_accuracy: 0.8762
Epoch 36/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.2204 - accuracy: 0.9241 - val_loss: 0.4246 - val_accuracy: 0.8769
Epoch 37/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.2150 - accuracy: 0.9251 - val_loss: 0.3939 - val_accuracy: 0.8797
Epoch 38/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.2156 - accuracy: 0.9261 - val_loss: 0.4005 - val_accuracy: 0.8790
Epoch 39/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.2135 - accuracy: 0.9268 - val_loss: 0.3959 - val_accuracy: 0.8773
Epoch 40/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.2121 - accuracy: 0.9263 - val_loss: 0.4072 - val_accuracy: 0.8742
Epoch 41/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.2087 - accuracy: 0.9277 - val_loss: 0.4234 - val_accuracy: 0.8769
Epoch 42/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.2023 - accuracy: 0.9310 - val_loss: 0.3904 - val_accuracy: 0.8812
Epoch 43/50
1562/1562 [==============================] - 26s 16ms/step - loss: 0.1986 - accuracy: 0.9311 - val_loss: 0.3859 - val_accuracy: 0.8806
Epoch 44/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.1936 - accuracy: 0.9341 - val_loss: 0.4627 - val_accuracy: 0.8703
Epoch 45/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.1907 - accuracy: 0.9342 - val_loss: 0.4460 - val_accuracy: 0.8646
Epoch 46/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.1884 - accuracy: 0.9341 - val_loss: 0.4511 - val_accuracy: 0.8658
Epoch 47/50
1562/1562 [==============================] - 26s 17ms/step - loss: 0.1877 - accuracy: 0.9344 - val_loss: 0.3790 - val_accuracy: 0.8831
Epoch 48/50
1562/1562 [==============================] - 26s 16ms/step - loss: 0.1851 - accuracy: 0.9366 - val_loss: 0.4208 - val_accuracy: 0.8770
Epoch 49/50
1562/1562 [==============================] - 26s 16ms/step - loss: 0.1869 - accuracy: 0.9364 - val_loss: 0.3946 - val_accuracy: 0.8841
Epoch 50/50
1562/1562 [==============================] - 26s 16ms/step - loss: 0.1767 - accuracy: 0.9397 - val_loss: 0.4432 - val_accuracy: 0.8781
# Plot loss per iteration
import matplotlib.pyplot as plt
plt.plot(r.history['loss'], label='loss')
plt.plot(r.history['val_loss'], label='val_loss')
plt.legend()
<matplotlib.legend.Legend at 0x7f1d5a2856d8>
# Plot accuracy per iteration
plt.plot(r.history['accuracy'], label='acc')
plt.plot(r.history['val_accuracy'], label='val_acc')
plt.legend()
<matplotlib.legend.Legend at 0x7f1d5a26ada0>
# Plot confusion matrix
from sklearn.metrics import confusion_matrix
import itertools
def plot_confusion_matrix(cm, classes,
normalize=False,
title='Confusion matrix',
cmap=plt.cm.Blues):
"""
This function prints and plots the confusion matrix.
Normalization can be applied by setting `normalize=True`.
"""
if normalize:
cm = cm.astype('float') / cm.sum(axis=1)[:, np.newaxis]
print("Normalized confusion matrix")
else:
print('Confusion matrix, without normalization')
print(cm)
plt.imshow(cm, interpolation='nearest', cmap=cmap)
plt.title(title)
plt.colorbar()
tick_marks = np.arange(len(classes))
plt.xticks(tick_marks, classes, rotation=45)
plt.yticks(tick_marks, classes)
fmt = '.2f' if normalize else 'd'
thresh = cm.max() / 2.
for i, j in itertools.product(range(cm.shape[0]), range(cm.shape[1])):
plt.text(j, i, format(cm[i, j], fmt),
horizontalalignment="center",
color="white" if cm[i, j] > thresh else "black")
plt.tight_layout()
plt.ylabel('True label')
plt.xlabel('Predicted label')
plt.show()
p_test = model.predict(x_test).argmax(axis=1)
cm = confusion_matrix(y_test, p_test)
plot_confusion_matrix(cm, list(range(10)))
Confusion matrix, without normalization
[[908 14 12 2 1 0 2 2 33 26]
[ 5 957 0 0 0 1 3 0 4 30]
[ 42 4 817 18 28 18 43 11 7 12]
[ 22 18 27 742 29 62 48 19 13 20]
[ 16 2 39 22 847 16 34 15 5 4]
[ 5 10 22 94 23 790 21 23 4 8]
[ 4 2 16 18 4 2 946 1 4 3]
[ 15 5 6 15 27 6 5 908 6 7]
[ 37 12 1 3 0 0 1 2 921 23]
[ 9 37 0 0 1 1 2 1 4 945]]
# Show some misclassified examples
misclassified_idx = np.where(p_test != y_test)[0]
i = np.random.choice(misclassified_idx)
plt.imshow(x_test[i], cmap='gray')
plt.title("True label: %s Predicted: %s" % (labels[y_test[i]], labels[p_test[i]]));
Model: "model"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
input_1 (InputLayer) [(None, 32, 32, 3)] 0
_________________________________________________________________
conv2d (Conv2D) (None, 32, 32, 32) 896
_________________________________________________________________
batch_normalization (BatchNo (None, 32, 32, 32) 128
_________________________________________________________________
conv2d_1 (Conv2D) (None, 32, 32, 32) 9248
_________________________________________________________________
batch_normalization_1 (Batch (None, 32, 32, 32) 128
_________________________________________________________________
max_pooling2d (MaxPooling2D) (None, 16, 16, 32) 0
_________________________________________________________________
conv2d_2 (Conv2D) (None, 16, 16, 64) 18496
_________________________________________________________________
batch_normalization_2 (Batch (None, 16, 16, 64) 256
_________________________________________________________________
conv2d_3 (Conv2D) (None, 16, 16, 64) 36928
_________________________________________________________________
batch_normalization_3 (Batch (None, 16, 16, 64) 256
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 8, 8, 64) 0
_________________________________________________________________
conv2d_4 (Conv2D) (None, 8, 8, 128) 73856
_________________________________________________________________
batch_normalization_4 (Batch (None, 8, 8, 128) 512
_________________________________________________________________
conv2d_5 (Conv2D) (None, 8, 8, 128) 147584
_________________________________________________________________
batch_normalization_5 (Batch (None, 8, 8, 128) 512
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 4, 4, 128) 0
_________________________________________________________________
flatten (Flatten) (None, 2048) 0
_________________________________________________________________
dropout (Dropout) (None, 2048) 0
_________________________________________________________________
dense (Dense) (None, 1024) 2098176
_________________________________________________________________
dropout_1 (Dropout) (None, 1024) 0
_________________________________________________________________
dense_1 (Dense) (None, 10) 10250
=================================================================
Total params: 2,397,226
Trainable params: 2,396,330
Non-trainable params: 896
_________________________________________________________________
