keras訓練淺層卷積網路並儲存和載入模型
阿新 • • 發佈：2018-12-29
這裡我們使用keras定義簡單的神經網路全連線層訓練MNIST資料集和cifar10資料集：
keras_mnist.py
from sklearn.preprocessing import LabelBinarizer from sklearn.model_selection import train_test_split from sklearn.metrics import classification_report from keras.models import Sequential from keras.layers.core import Dense from keras.optimizers import SGD from sklearn import datasets import matplotlib.pyplot as plt import numpy as np import argparse # 命令列引數執行 ap = argparse.ArgumentParser() ap.add_argument("-o", "--output", required=True, help="path to the output loss/accuracy plot") args =vars(ap.parse_args()) # 載入資料MNIST，然後歸一化到【0,1】，同時使用75%做訓練，25%做測試 print("[INFO] loading MNIST (full) dataset") dataset = datasets.fetch_mldata("MNIST Original", data_home="/home/king/test/python/train/pyimagesearch/nn/data/") data = dataset.data.astype("float") / 255.0 (trainX, testX, trainY, testY) = train_test_split(data, dataset.target, test_size=0.25) # 將label進行one-hot編碼 lb = LabelBinarizer() trainY = lb.fit_transform(trainY) testY = lb.transform(testY) # keras定義網路結構784--256--128--10 model = Sequential() model.add(Dense(256, input_shape=(784,), activation="relu")) model.add(Dense(128, activation="relu")) model.add(Dense(10, activation="softmax")) # 開始訓練 print("[INFO] training network...") # 0.01的學習率 sgd = SGD(0.01) # 交叉驗證 model.compile(loss="categorical_crossentropy", optimizer=sgd, metrics=['accuracy']) H = model.fit(trainX, trainY, validation_data=(testX, testY), epochs=100, batch_size=128) # 測試模型和評估 print("[INFO] evaluating network...") predictions = model.predict(testX, batch_size=128) print(classification_report(testY.argmax(axis=1), predictions.argmax(axis=1), target_names=[str(x) for x in lb.classes_])) # 儲存視覺化訓練結果 plt.style.use("ggplot") plt.figure() plt.plot(np.arange(0, 100), H.history["loss"], label="train_loss") plt.plot(np.arange(0, 100), H.history["val_loss"], label="val_loss") plt.plot(np.arange(0, 100), H.history["acc"], label="train_acc") plt.plot(np.arange(0, 100), H.history["val_acc"], label="val_acc") plt.title("Training Loss and Accuracy") plt.xlabel("# Epoch") plt.ylabel("Loss/Accuracy") plt.legend() plt.savefig(args["output"])
使用relu做啟用函式：
使用sigmoid做啟用函式：
接著我們自己定義一些modules去實現一個簡單的卷基層去訓練cifar10資料集：
imagetoarraypreprocessor.py
''' 該函式主要是實現keras的一個細節轉換，因為訓練的影象時RGB三顏色通道，讀取進來的資料是有depth的，keras為了相容一些後臺，預設是按照（height， width， depth）讀取，但有時候就要改變成（depth， height， width） ''' from keras.preprocessing.image import img_to_array class ImageToArrayPreprocessor: def __init__(self, dataFormat=None): self.dataFormat = dataFormat def preprocess(self, image): return img_to_array(image, data_format=self.dataFormat)
shallownet.py
''' 定義一個簡單的卷基層： input->conv->Relu->FC ''' from keras.models import Sequential from keras.layers.convolutional import Conv2D from keras.layers.core import Activation, Flatten, Dense from keras import backend as K class ShallowNet: @staticmethod def build(width, height, depth, classes): model = Sequential() inputShape = (height, width, depth) if K.image_data_format() == "channels_first": inputShape = (depth, height, width) model.add(Conv2D(32, (3, 3), padding="same", input_shape=inputShape)) model.add(Activation("relu")) model.add(Flatten()) model.add(Dense(classes)) model.add(Activation("softmax")) return model
然後就是訓練程式碼：
keras_cifar10.py
from sklearn.preprocessing import LabelBinarizer
from sklearn.metrics import classification_report
from shallownet import ShallowNet
from keras.optimizers import SGD
from keras.datasets import cifar10
import matplotlib.pyplot as plt
import numpy as np
import argparse
ap = argparse.ArgumentParser()
ap.add_argument("-o", "--output", required=True, help="path to the output loss/accuracy plot")
args = vars(ap.parse_args())
print("[INFO] loading CIFAR-10 dataset")
((trainX, trainY), (testX, testY)) = cifar10.load_data()
trainX = trainX.astype("float") / 255.0
testX = testX.astype("float") / 255.0
lb = LabelBinarizer()
trainY = lb.fit_transform(trainY)
testY = lb.transform(testY)
# 標籤0-9代表的類別string
labelNames = ['airplane', 'automobile', 'bird', 'cat',
'deer', 'dog', 'frog', 'horse', 'ship', 'truck']
print("[INFO] compiling model...")
opt = SGD(lr=0.0001)
model = ShallowNet.build(width=32, height=32, depth=3, classes=10)
model.compile(loss="categorical_crossentropy", optimizer=opt, metrics=["accuracy"])
print("[INFO] training network...")
H = model.fit(trainX, trainY, validation_data=(testX, testY), batch_size=32, epochs=1000, verbose=1)
print("[INFO] evaluating network...")
predictions = model.predict(testX, batch_size=32)
print(classification_report(testY.argmax(axis=1), predictions.argmax(axis=1),
target_names=labelNames))
# 儲存視覺化訓練結果
plt.style.use("ggplot")
plt.figure()
plt.plot(np.arange(0, 1000), H.history["loss"], label="train_loss")
plt.plot(np.arange(0, 1000), H.history["val_loss"], label="val_loss")
plt.plot(np.arange(0, 1000), H.history["acc"], label="train_acc")
plt.plot(np.arange(0, 1000), H.history["val_acc"], label="val_acc")
plt.title("Training Loss and Accuracy")
plt.xlabel("# Epoch")
plt.ylabel("Loss/Accuracy")
plt.legend()
plt.savefig(args["output"])
程式碼中可以對訓練的learning rate進行微調，大概可以接近60%的準確率。
然後修改下程式碼可以儲存訓練模型：
from sklearn.preprocessing import LabelBinarizer
from sklearn.metrics import classification_report
from shallownet import ShallowNet
from keras.optimizers import SGD
from keras.datasets import cifar10
import matplotlib.pyplot as plt
import numpy as np
import argparse
ap = argparse.ArgumentParser()
ap.add_argument("-o", "--output", required=True, help="path to the output loss/accuracy plot")
ap.add_argument("-m", "--model", required=True, help="path to save train model")
args = vars(ap.parse_args())
print("[INFO] loading CIFAR-10 dataset")
((trainX, trainY), (testX, testY)) = cifar10.load_data()
trainX = trainX.astype("float") / 255.0
testX = testX.astype("float") / 255.0
lb = LabelBinarizer()
trainY = lb.fit_transform(trainY)
testY = lb.transform(testY)
# 標籤0-9代表的類別string
labelNames = ['airplane', 'automobile', 'bird', 'cat',
'deer', 'dog', 'frog', 'horse', 'ship', 'truck']
print("[INFO] compiling model...")
opt = SGD(lr=0.005)
model = ShallowNet.build(width=32, height=32, depth=3, classes=10)
model.compile(loss="categorical_crossentropy", optimizer=opt, metrics=["accuracy"])
print("[INFO] training network...")
H = model.fit(trainX, trainY, validation_data=(testX, testY), batch_size=32, epochs=50, verbose=1)
model.save(args["model"])
print("[INFO] evaluating network...")
predictions = model.predict(testX, batch_size=32)
print(classification_report(testY.argmax(axis=1), predictions.argmax(axis=1),
target_names=labelNames))
# 儲存視覺化訓練結果
plt.style.use("ggplot")
plt.figure()
plt.plot(np.arange(0, 5), H.history["loss"], label="train_loss")
plt.plot(np.arange(0, 5), H.history["val_loss"], label="val_loss")
plt.plot(np.arange(0, 5), H.history["acc"], label="train_acc")
plt.plot(np.arange(0, 5), H.history["val_acc"], label="val_acc")
plt.title("Training Loss and Accuracy")
plt.xlabel("# Epoch")
plt.ylabel("Loss/Accuracy")
plt.legend()
plt.savefig(args["output"])
命令列執行：
我們使用另一個程式來載入上一次訓練儲存的模型，然後進行測試：
test.py
from sklearn.preprocessing import LabelBinarizer
from sklearn.metrics import classification_report
from shallownet import ShallowNet
from keras.optimizers import SGD
from keras.datasets import cifar10
from keras.models import load_model
import matplotlib.pyplot as plt
import numpy as np
import argparse
ap = argparse.ArgumentParser()
ap.add_argument("-m", "--model", required=True, help="path to save train model")
args = vars(ap.parse_args())
# 標籤0-9代表的類別string
labelNames = ['airplane', 'automobile', 'bird', 'cat',
'deer', 'dog', 'frog', 'horse', 'ship', 'truck']
print("[INFO] loading CIFAR-10 dataset")
((trainX, trainY), (testX, testY)) = cifar10.load_data()
idxs = np.random.randint(0, len(testX), size=(10,))
testX = testX[idxs]
testY = testY[idxs]
trainX = trainX.astype("float") / 255.0
testX = testX.astype("float") / 255.0
lb = LabelBinarizer()
trainY = lb.fit_transform(trainY)
testY = lb.transform(testY)
print("[INFO] loading pre-trained network...")
model = load_model(args["model"])
print("[INFO] evaluating network...")
predictions = model.predict(testX, batch_size=32).argmax(axis=1)
print("predictions\n", predictions)
for i in range(len(testY)):
print("label:{}".format(labelNames[predictions[i]]))
trueLabel = []
for i in range(len(testY)):
for j in range(len(testY[i])):
if testY[i][j] != 0:
trueLabel.append(j)
print(trueLabel)
print("ground truth testY:")
for i in range(len(trueLabel)):
print("label:{}".format(labelNames[trueLabel[i]]))
print("TestY\n", testY)
ok