Transfer learning¶
Main intuition:¶
- Train a model using a specific set of training data
- Save the model
- Use the pre-trained model for a new problem/dataset
- Re-train using the new dataset
When do we use it:¶
If we have a small dataset, not big enough for proper training. The use of a pre-trained model on much larger dataset provides a better initialization
Examples:¶
The most common example is: ImageNet network + transfer learning to specific dataset
In [0]:
from __future__ import print_function
!pip install q keras==2.3.1
import datetime
import keras
from keras.datasets import mnist
from keras.models import Sequential
from keras.layers import Dense, Dropout, Activation, Flatten
from keras.layers import Conv2D, MaxPooling2D
from keras import backend as K
In [0]:
now = datetime.datetime.now
batch_size = 128
num_classes = 5
epochs = 5
# input image dimensions
img_rows, img_cols = 28, 28
# number of convolutional filters to use
filters = 32
# size of pooling area for max pooling
pool_size = 2
# convolution kernel size
kernel_size = 3
In [0]:
K.backend()
Out[0]:
'tensorflow'
In [0]:
if K.image_data_format() == 'channels_first':
input_shape = (1, img_rows, img_cols)
else:
input_shape = (img_rows, img_cols, 1)
In [0]:
def train_model(model, train, test, num_classes):
x_train = train[0].reshape((train[0].shape[0],) + input_shape)
x_test = test[0].reshape((test[0].shape[0],) + input_shape)
x_train = x_train.astype('float32')
x_test = x_test.astype('float32')
x_train /= 255
x_test /= 255
print('x_train shape:', x_train.shape)
print(x_train.shape[0], 'train samples')
print(x_test.shape[0], 'test samples')
# convert class vectors to binary class matrices
y_train = keras.utils.to_categorical(train[1], num_classes)
y_test = keras.utils.to_categorical(test[1], num_classes)
model.compile(loss='categorical_crossentropy',
optimizer='adadelta',
metrics=['accuracy'])
t = now()
model.fit(x_train, y_train,
batch_size=batch_size,
epochs=epochs,
verbose=1,
validation_data=(x_test, y_test))
print('Training time: %s' % (now() - t))
score = model.evaluate(x_test, y_test, verbose=0)
print('Test score:', score[0])
print('Test accuracy:', score[1])
In [0]:
# the data, split between train and test sets
(x_train, y_train), (x_test, y_test) = mnist.load_data()
In [0]:
# create two datasets one with digits below 5 and one with 5 and above
x_train_lt5 = x_train[y_train < 5]
y_train_lt5 = y_train[y_train < 5]
x_test_lt5 = x_test[y_test < 5]
y_test_lt5 = y_test[y_test < 5]
x_train_gte5 = x_train[y_train >= 5]
y_train_gte5 = y_train[y_train >= 5] - 5
x_test_gte5 = x_test[y_test >= 5]
y_test_gte5 = y_test[y_test >= 5] - 5
In [0]:
# define two groups of layers: feature (convolutions) and classification (dense)
feature_layers = [
Conv2D(filters, kernel_size,
padding='valid',
input_shape=input_shape),
Activation('relu'),
Conv2D(filters, kernel_size),
Activation('relu'),
MaxPooling2D(pool_size=pool_size),
Dropout(0.25),
Flatten(),
]
classification_layers = [
Dense(128),
Activation('relu'),
Dropout(0.5),
Dense(num_classes),
Activation('softmax')
]
In [0]:
# create complete model
model = Sequential(feature_layers + classification_layers)
In [0]:
from keras.utils import plot_model
#plot_model(model, to_file='model_plot.png', show_shapes=True, show_layer_names=True)
plot_model(model, show_shapes=True, show_layer_names=True)
Out[0]:
In [0]:
# train model for 5-digit classification [0..4]
train_model(model,
(x_train_lt5, y_train_lt5),
(x_test_lt5, y_test_lt5), num_classes)
x_train shape: (30596, 28, 28, 1) 30596 train samples 5139 test samples Train on 30596 samples, validate on 5139 samples Epoch 1/5 30596/30596 [==============================] - 49s 2ms/step - loss: 0.1637 - accuracy: 0.9488 - val_loss: 0.0325 - val_accuracy: 0.9905 Epoch 2/5 30596/30596 [==============================] - 49s 2ms/step - loss: 0.0491 - accuracy: 0.9860 - val_loss: 0.0159 - val_accuracy: 0.9944 Epoch 3/5 30596/30596 [==============================] - 48s 2ms/step - loss: 0.0319 - accuracy: 0.9907 - val_loss: 0.0134 - val_accuracy: 0.9944 Epoch 4/5 30596/30596 [==============================] - 48s 2ms/step - loss: 0.0272 - accuracy: 0.9917 - val_loss: 0.0078 - val_accuracy: 0.9971 Epoch 5/5 30596/30596 [==============================] - 48s 2ms/step - loss: 0.0221 - accuracy: 0.9930 - val_loss: 0.0083 - val_accuracy: 0.9967 Training time: 0:04:03.773571 Test score: 0.008321681414331934 Test accuracy: 0.9966919422149658
In [0]:
# freeze feature layers and rebuild model
for l in feature_layers:
l.trainable = False
In [0]:
# transfer: train dense layers for new classification task [5..9]
train_model(model,
(x_train_gte5, y_train_gte5),
(x_test_gte5, y_test_gte5), num_classes)
x_train shape: (29404, 28, 28, 1) 29404 train samples 4861 test samples Train on 29404 samples, validate on 4861 samples Epoch 1/5 29404/29404 [==============================] - 17s 564us/step - loss: 0.0420 - accuracy: 0.9869 - val_loss: 0.0250 - val_accuracy: 0.9912 Epoch 2/5 29404/29404 [==============================] - 16s 556us/step - loss: 0.0394 - accuracy: 0.9884 - val_loss: 0.0227 - val_accuracy: 0.9934 Epoch 3/5 29404/29404 [==============================] - 16s 560us/step - loss: 0.0346 - accuracy: 0.9895 - val_loss: 0.0235 - val_accuracy: 0.9926 Epoch 4/5 29404/29404 [==============================] - 16s 559us/step - loss: 0.0337 - accuracy: 0.9892 - val_loss: 0.0233 - val_accuracy: 0.9924 Epoch 5/5 29404/29404 [==============================] - 16s 559us/step - loss: 0.0328 - accuracy: 0.9904 - val_loss: 0.0210 - val_accuracy: 0.9934 Training time: 0:01:22.717211 Test score: 0.02103791343513074 Test accuracy: 0.993416965007782