CIFAR10 是另外一個 dataset， 和 mnist 一樣，有十種類別（飛機、汽車、鳥、貓、鹿、狗、青蛙、馬、船、卡車）
In [1]:
import keras
from keras.models import Sequential
from PIL import Image
import numpy as np
import tarfile
In [2]:
# 讀取 dataset
# 只有 train 和 test 沒有 validation
import pickle
train_X=[]
train_y=[]
tar_gz = "../Week06/cifar-10-python.tar.gz"
with tarfile.open(tar_gz) as tarf:
for i in range(1, 6):
dataset = "cifar-10-batches-py/data_batch_%d"%i
print("load",dataset)
with tarf.extractfile(dataset) as f:
result = pickle.load(f, encoding='latin1')
train_X.extend(result['data']/255)
train_y.extend(result['labels'])
train_X=np.float32(train_X)
train_y=np.int32(train_y)
dataset = "cifar-10-batches-py/test_batch"
print("load",dataset)
with tarf.extractfile(dataset) as f:
result = pickle.load(f, encoding='latin1')
test_X=np.float32(result['data']/255)
test_y=np.int32(result['labels'])
train_Y = np.eye(10)[train_y]
test_Y = np.eye(10)[test_y]
validation_data = (test_X[:1000], test_Y[:1000])
test_data = (test_X[1000:], test_Y[1000:])
In [3]:
from IPython.display import display
def showX(X):
int_X = (X*255).clip(0,255).astype('uint8')
# N*3072 -> N*3*32*32 -> 32 * 32N * 3
int_X_reshape = np.moveaxis(int_X.reshape(-1,3,32,32), 1, 3)
int_X_reshape = int_X_reshape.swapaxes(0,1).reshape(32,-1, 3)
display(Image.fromarray(int_X_reshape))
# 訓練資料， X 的前 20 筆
showX(train_X[:20])
print(train_y[:20])
name_array = np.array("飛機、汽車、鳥、貓、鹿、狗、青蛙、馬、船、卡車".split('、'))
print(name_array[train_y[:20]])
將之前的 cnn model 套用過來看看
In [4]:
# %load ../Week06/q_cifar10_cnn.py
import keras
from keras.layers import Dense, Activation, Conv2D, MaxPool2D, Reshape
model = Sequential()
model.add(Reshape((3, 32, 32), input_shape=(3*32*32,) ))
model.add(Conv2D(filters=32, kernel_size=(3,3), padding='same', activation="relu", data_format='channels_first'))
model.add(MaxPool2D())
model.add(Conv2D(filters=64, kernel_size=(3,3), padding='same', activation="relu", data_format='channels_first'))
model.add(MaxPool2D())
model.add(Reshape((-1,)))
model.add(Dense(units=1024, activation="relu"))
model.add(Dense(units=10, activation="softmax"))
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
model.fit(train_X, train_Y, validation_data=validation_data, batch_size=100, epochs=10)
rtn = model.evaluate(*test_data)
print("\ntest accuracy=", rtn[1])
In [5]:
showX(test_X[:15])
predict_y = model.predict_classes(test_X[:15], verbose=False)
print(name_array[predict_y])
print(name_array[test_y[:15]])
In [6]:
import keras
from keras.layers import Dense, Activation, Conv2D, MaxPool2D, Reshape, Dropout
model = Sequential()
model.add(Reshape((3, 32, 32), input_shape=(3*32*32,) ))
model.add(Conv2D(32, 3, padding='same', activation="relu", data_format='channels_first'))
model.add(MaxPool2D())
model.add(Conv2D(64, 3, padding='same', activation="relu", data_format='channels_first'))
model.add(MaxPool2D())
model.add(Reshape((-1,)))
model.add(Dense(units=1024, activation="relu"))
model.add(Dropout(rate=0.4))
model.add(Dense(units=10, activation="softmax"))
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
model.fit(train_X, train_Y, validation_data=validation_data, batch_size=100, epochs=10)
rtn = model.evaluate(*test_data)
print("\ntest accuracy=", rtn[1])
In [7]:
model.fit(train_X, train_Y, validation_data=validation_data, batch_size=100, epochs=10)
rtn = model.evaluate(*test_data)
print("\ntest accuracy=", rtn[1])
In [8]:
showX(test_X[:15])
predict_y = model.predict_classes(test_X[:15], verbose=False)
print(name_array[predict_y])
print(name_array[test_y[:15]])
不同的 activation https://keras.io/activations/
In [4]:
# 先定義一個工具
def add_layers(model, *layers):
for l in layers:
model.add(l)
In [10]:
import keras
from keras.engine.topology import Layer
from keras.layers import Dense, Activation, Conv2D, MaxPool2D, Reshape, Dropout
def MyConv2D(filters, kernel_size, **kwargs):
return (Conv2D(filters=filters, kernel_size=kernel_size,
padding='same', data_format='channels_first', **kwargs),
Activation("elu"))
model = Sequential()
add_layers( model,
Reshape((3, 32, 32), input_shape=(3*32*32,)),
*MyConv2D(32, 3),
MaxPool2D(),
*MyConv2D(64, 3),
MaxPool2D(),
Reshape((-1,)),
Dense(units=1024, activation="elu"),
Dropout(rate=0.4),
Dense(units=10, activation="softmax")
)
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
model.fit(train_X, train_Y, validation_data=validation_data, batch_size=100, epochs=10)
rtn = model.evaluate(*test_data)
print("\ntest accuracy=", rtn[1])
In [11]:
import keras
from keras.layers import Dense, Activation, Conv2D, MaxPool2D, Reshape, Dropout, BatchNormalization
# 先處理資料
#GCN
train_X_mean = np.mean(train_X, axis=0, keepdims=True)
train_X_std = np.std(train_X, axis=0, keepdims=True)
preprocessed_train_X = (train_X-train_X_mean)/train_X_std
preprocessed_test_X = (test_X-train_X_mean)/train_X_std
preprocessed_validation_data = (preprocessed_test_X[:1000], test_Y[:1000])
preprocessed_test_data = (preprocessed_test_X[1000:], test_Y[1000:])
def MyConv2D(filters, kernel_size, **kwargs):
return (Conv2D(filters=filters, kernel_size=kernel_size,
padding='same', data_format='channels_first', **kwargs),
Activation("relu"))
def add_layers(model, *layers):
for l in layers:
model.add(l)
model = Sequential()
add_layers( model,
Reshape((3, 32, 32), input_shape=(3*32*32,)),
*MyConv2D(32, 3),
MaxPool2D(),
*MyConv2D(64, 3),
MaxPool2D(),
Reshape((-1,)),
Dense(units=1024, activation="relu"),
Dropout(rate=0.4),
Dense(units=10, activation="softmax")
)
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
model.fit(preprocessed_train_X, train_Y, validation_data=preprocessed_validation_data,
batch_size=100, epochs=10)
rtn = model.evaluate(*preprocessed_test_data)
print("\ntest accuracy=", rtn[1])
In [12]:
import keras
from keras.layers import Dense, Activation, Conv2D, MaxPool2D, Reshape, Dropout, BatchNormalization
# 先處理資料
#GCN
train_X_mean = np.mean(train_X, axis=0, keepdims=True)
train_X_std = np.std(train_X, axis=0, keepdims=True)
preprocessed_train_X = (train_X-train_X_mean)/train_X_std
preprocessed_test_X = (test_X-train_X_mean)/train_X_std
preprocessed_validation_data = (preprocessed_test_X[:1000], test_Y[:1000])
preprocessed_test_data = (preprocessed_test_X[1000:], test_Y[1000:])
def MyConv2D(filters, kernel_size, **kwargs):
return (Conv2D(filters=filters, kernel_size=kernel_size,
padding='same', data_format='channels_first', **kwargs),
BatchNormalization(axis=1),
Activation("relu"))
def add_layers(model, *layers):
for l in layers:
model.add(l)
model = Sequential()
add_layers( model,
Reshape((3, 32, 32), input_shape=(3*32*32,)),
*MyConv2D(32, 3),
MaxPool2D(),
*MyConv2D(64, 3),
MaxPool2D(),
Reshape((-1,)),
Dense(units=1024, activation="relu"),
Dropout(0.4),
Dense(units=10, activation="softmax")
)
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
model.fit(preprocessed_train_X, train_Y, validation_data=preprocessed_validation_data,
batch_size=100, epochs=10)
rtn = model.evaluate(*preprocessed_test_data)
print("\ntest accuracy=", rtn[1])
In [13]:
model.fit(preprocessed_train_X, train_Y, validation_data=preprocessed_validation_data,
batch_size=100, epochs=10)
rtn = model.evaluate(*preprocessed_test_data)
print("\ntest accuracy=", rtn[1])
In [14]:
def zca_whitening_matrix(X):
"""
Function to compute ZCA whitening matrix (aka Mahalanobis whitening).
INPUT: X: [M x N] matrix.
Rows: Variables
Columns: Observations
OUTPUT: ZCAMatrix: [M x M] matrix
"""
X = X.T
# Covariance matrix [column-wise variables]: Sigma = (X-mu)' * (X-mu) / N
sigma = np.cov(X, rowvar=True) # [M x M]
# Singular Value Decomposition. X = U * np.diag(S) * V
U,S,V = np.linalg.svd(sigma)
# U: [M x M] eigenvectors of sigma.
# S: [M x 1] eigenvalues of sigma.
# V: [M x M] transpose of U
# Whitening constant: prevents division by zero
epsilon = 1e-5
# ZCA Whitening matrix: U * Lambda * U'
ZCAMatrix = np.dot(U, np.dot(np.diag(1.0/np.sqrt(S + epsilon)), U.T)) # [M x M]
return ZCAMatrix.T
# ZCAMatrix = zca_whitening_matrix(X0)
# new_train_X= ((train_X-train_X_mean)/train_X_std) @ ZCAMatrix
In [15]:
# 參考 https://keras.io/preprocessing/image/
# 輸入改成 tensor4
train_X = train_X.reshape(-1, 3, 32, 32)
test_X = test_X.reshape(-1, 3, 32, 32)
def MyConv2D(filters, kernel_size, **kwargs):
return (Conv2D(filters=filters, kernel_size=kernel_size,
padding='same', data_format='channels_first', **kwargs),
BatchNormalization(axis=1),
Activation("relu"))
model = Sequential()
add_layers( model,
*MyConv2D(32, 3, input_shape=(3,32,32)),
MaxPool2D(),
*MyConv2D(64, 3),
MaxPool2D(),
Reshape((-1,)),
Dense(units=1024, activation="relu"),
Dropout(0.4),
Dense(units=10, activation="softmax")
)
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
# 使用 keras 的功能
from keras.preprocessing.image import ImageDataGenerator
datagen = ImageDataGenerator(
featurewise_center=True,
featurewise_std_normalization=True,
zca_whitening=True,
data_format="channels_first")
# compute quantities required for featurewise normalization
# (std, mean, and principal components if ZCA whitening is applied)
datagen.fit(train_X)
p_train_X, p_train_Y = datagen.flow(train_X, train_Y, batch_size=len(train_X), shuffle=False).next()
# 順序都沒變
assert (p_train_Y == train_Y).all()
p_test_X, p_test_Y = datagen.flow(test_X, test_Y, batch_size=len(test_X), shuffle=False).next()
# 順序都沒變
assert (p_test_Y == test_Y).all()
# 不需要這兩個
del p_train_Y, p_test_Y
p_validation_data = (p_test_X[:1000], test_Y[:1000])
p_test_data = (p_test_X[1000:], test_Y[1000:])
model.fit(p_train_X, train_Y, validation_data=p_validation_data,
batch_size=100, epochs=10)
rtn = model.evaluate(*p_test_data)
print("\ntest accuracy=", rtn[1])
使用動態資料處理
# fits the model on batches with real-time data augmentation:
train_generator = datagen.flow(train_X, train_Y, batch_size=100, shuffle=False)
test_generator = datagen.flow(*test_data, batch_size=100, shuffle=False)
model.fit_generator(train_generator,
steps_per_epoch=len(train_X),
#validation_data=datagen.flow(*validation_data, batch_size=100),
#validation_steps=1000,
epochs=10)
rtn = model.evaluate_generator(test_generator, steps=9000)
In [16]:
import keras
from keras.layers import Dense, Activation, Conv2D, MaxPool2D, Reshape, Dropout, BatchNormalization
# 輸入改成 tensor4
train_X = train_X.reshape(-1, 3, 32, 32)
test_X = test_X.reshape(-1, 3, 32, 32)
def MyConv2D(filters, kernel_size, **kwargs):
return (Conv2D(filters=filters, kernel_size=kernel_size,
padding='same', data_format='channels_first', **kwargs),
BatchNormalization(axis=1),
Activation("elu"))
model = Sequential()
add_layers( model,
*MyConv2D(64, 3, input_shape=(3,32,32)),
*MyConv2D(64, 3),
MaxPool2D(),
*MyConv2D(128, 3),
*MyConv2D(128, 3),
MaxPool2D(),
*MyConv2D(256, 3),
*MyConv2D(256, 3),
Reshape((-1,)),
Dense(units=1024),
BatchNormalization(),
Activation("elu"),
Dropout(0.4),
Dense(units=10, activation="softmax")
)
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
# 使用 keras 的功能
from keras.preprocessing.image import ImageDataGenerator
datagen = ImageDataGenerator(
featurewise_center=True,
featurewise_std_normalization=True,
zca_whitening=True,
data_format="channels_first")
# compute quantities required for featurewise normalization
# (std, mean, and principal components if ZCA whitening is applied)
datagen.fit(train_X)
p_train_X, p_train_Y = datagen.flow(train_X, train_Y, batch_size=len(train_X), shuffle=False).next()
# 順序都沒變
assert (p_train_Y == train_Y).all()
p_test_X, p_test_Y = datagen.flow(test_X, test_Y, batch_size=len(test_X), shuffle=False).next()
# 順序都沒變
assert (p_test_Y == test_Y).all()
# 不需要這兩個
del p_train_Y, p_test_Y
p_validation_data = (p_test_X[:1000], test_Y[:1000])
p_test_data = (p_test_X[1000:], test_Y[1000:])
model.fit(p_train_X, train_Y, validation_data=p_validation_data,
batch_size=100, epochs=10)
rtn = model.evaluate(*p_test_data)
print("\ntest accuracy=", rtn[1])
In [ ]:
model.fit(p_train_X, train_Y, validation_data=p_validation_data,
batch_size=100, epochs=10)
rtn = model.evaluate(*p_test_data)
lasagne 中相同的
_ = InputLayer(shape=(None, 3*32*32), input_var=input_var)
_ = DropoutLayer(_, 0.2)
_ = ReshapeLayer(_, ([0], 3, 32, 32))
_ = conv(_, 96, 3)
_ = conv(_, 96, 3)
_ = MaxPool2DDNNLayer(_, 3, 2)
_ = DropoutLayer(_, 0.5)
_ = conv(_, 192, 3)
_ = conv(_, 192, 3)
_ = MaxPool2DDNNLayer(_, 3, 2)
_ = DropoutLayer(_, 0.5)
_ = conv(_, 192, 3)
_ = conv(_, 192, 1)
_ = conv(_, 10, 1)
_ = Pool2DDNNLayer(_, 7, mode='average_exc_pad')
_ = FlattenLayer(_)
l_out = NonlinearityLayer(_, nonlinearity=lasagne.nonlinearities.softmax)
In [ ]:
import keras
from keras.layers import Dense, Activation, Conv2D, MaxPool2D, Reshape, Dropout, BatchNormalization, GlobalAveragePooling2D
# 輸入改成 tensor4
train_X = train_X.reshape(-1, 3, 32, 32)
test_X = test_X.reshape(-1, 3, 32, 32)
def MyConv2D(filters, kernel_size, **kwargs):
return (Conv2D(filters=filters, kernel_size=kernel_size,
padding='same', data_format='channels_first', **kwargs),
BatchNormalization(axis=1, momentum=0.9),
Activation("relu"))
model = Sequential()
add_layers( model,
Dropout(0.2, input_shape=(3,32,32)),
*MyConv2D(96, 3),
*MyConv2D(96, 3),
MaxPool2D(3, 2),
*MyConv2D(192, 3),
*MyConv2D(192, 3),
MaxPool2D(3, 2),
Dropout(0.5),
*MyConv2D(192, 3),
*MyConv2D(192, 1),
*MyConv2D(10, 1),
GlobalAveragePooling2D(data_format='channels_first'),
Activation("softmax")
)
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
model.fit(p_train_X, train_Y, validation_data=p_validation_data,
batch_size=100, epochs=50)
rtn = model.evaluate(*p_test_data)
print("\ntest accuracy=", rtn[1])
In [5]:
import keras
from keras.layers import Dense, Activation, Conv2D, MaxPool2D, Reshape, Dropout, BatchNormalization, GlobalAveragePooling2D
# 輸入改成 tensor4
train_X = train_X.reshape(-1, 3, 32, 32)
test_X = test_X.reshape(-1, 3, 32, 32)
def MyConv2D(filters, kernel_size, **kwargs):
return (Conv2D(filters=filters, kernel_size=kernel_size,
padding='same', data_format='channels_first', **kwargs),
BatchNormalization(axis=1, momentum=0.9),
Activation("relu"))
model = Sequential()
add_layers( model,
Dropout(0.2, input_shape=(3,32,32)),
*MyConv2D(96, 3),
*MyConv2D(96, 3),
MaxPool2D(3, 2),
*MyConv2D(192, 3),
*MyConv2D(192, 3),
MaxPool2D(3, 2),
Dropout(0.5),
*MyConv2D(192, 3),
*MyConv2D(192, 1),
*MyConv2D(10, 1),
GlobalAveragePooling2D(data_format='channels_first'),
Activation("softmax")
)
model.compile(loss='categorical_crossentropy',
optimizer='adam',
metrics=['accuracy'])
# 使用 keras 的功能
from keras.preprocessing.image import ImageDataGenerator
datagen = ImageDataGenerator(
featurewise_center=True,
featurewise_std_normalization=True,
zca_whitening=True,
data_format="channels_first")
# compute quantities required for featurewise normalization
# (std, mean, and principal components if ZCA whitening is applied)
datagen.fit(train_X)
p_train_X, p_train_Y = datagen.flow(train_X, train_Y, batch_size=len(train_X), shuffle=False).next()
# 順序都沒變
assert (p_train_Y == train_Y).all()
p_test_X, p_test_Y = datagen.flow(test_X, test_Y, batch_size=len(test_X), shuffle=False).next()
# 順序都沒變
assert (p_test_Y == test_Y).all()
# 不需要這兩個
del p_train_Y, p_test_Y
p_validation_data = (p_test_X[:1000], test_Y[:1000])
p_test_data = (p_test_X[1000:], test_Y[1000:])
model.fit(p_train_X, train_Y, validation_data=p_validation_data,
batch_size=100, epochs=10)
rtn = model.evaluate(*p_test_data)
print("\ntest accuracy=", rtn[1])
In [ ]: