Python實現(xiàn)Keras搭建神經(jīng)網(wǎng)絡訓練分類模型教程
更新時間：2020年06月12日 15:59:54 作者：Wish_97
這篇文章主要介紹了Python實現(xiàn)Keras搭建神經(jīng)網(wǎng)絡訓練分類模型教程，具有很好的參考價值，希望對大家有所幫助。一起跟隨小編過來看看吧
我就廢話不多說了，大家還是直接看代碼吧~
注釋講解版：
# Classifier example
import numpy as np
# for reproducibility
np.random.seed(1337)
# from keras.datasets import mnist
from keras.utils import np_utils
from keras.models import Sequential
from keras.layers import Dense, Activation
from keras.optimizers import RMSprop
# 程序中用到的數(shù)據(jù)是經(jīng)典的手寫體識別mnist數(shù)據(jù)集
# download the mnist to the path if it is the first time to be called
# X shape (60,000 28x28), y
# (X_train, y_train), (X_test, y_test) = mnist.load_data()
# 下載minst.npz：
# 鏈接: https://pan.baidu.com/s/1b2ppKDOdzDJxivgmyOoQsA
# 提取碼: y5ir
# 將下載好的minst.npz放到當前目錄下
path='./mnist.npz'
f = np.load(path)
X_train, y_train = f['x_train'], f['y_train']
X_test, y_test = f['x_test'], f['y_test']
f.close()
# data pre-processing
# 數(shù)據(jù)預處理
# normalize
# X shape (60,000 28x28)，表示輸入數(shù)據(jù) X 是個三維的數(shù)據(jù)
# 可以理解為 60000行數(shù)據(jù)，每一行是一張28 x 28 的灰度圖片
# X_train.reshape(X_train.shape[0], -1)表示：只保留第一維，其余的緯度，不管多少緯度，重新排列為一維
# 參數(shù)-1就是不知道行數(shù)或者列數(shù)多少的情況下使用的參數(shù)
# 所以先確定除了參數(shù)-1之外的其他參數(shù)，然后通過(總參數(shù)的計算) / (確定除了參數(shù)-1之外的其他參數(shù)) = 該位置應該是多少的參數(shù)
# 這里用-1是偷懶的做法,等同于 28*28
# reshape后的數(shù)據(jù)是：共60000行，每一行是784個數(shù)據(jù)點（feature）
# 輸入的 x 變成 60,000*784 的數(shù)據(jù)，然后除以 255 進行標準化
# 因為每個像素都是在 0 到 255 之間的，標準化之后就變成了 0 到 1 之間
X_train = X_train.reshape(X_train.shape[0], -1) / 255
X_test = X_test.reshape(X_test.shape[0], -1) / 255
# 分類標簽編碼
# 將y轉(zhuǎn)化為one-hot vector
y_train = np_utils.to_categorical(y_train, num_classes = 10)
y_test = np_utils.to_categorical(y_test, num_classes = 10)
# Another way to build your neural net
# 建立神經(jīng)網(wǎng)絡
# 應用了2層的神經(jīng)網(wǎng)絡，前一層的激活函數(shù)用的是relu，后一層的激活函數(shù)用的是softmax
#32是輸出的維數(shù)
model = Sequential([
Dense(32, input_dim=784),
Activation('relu'),
Dense(10),
Activation('softmax')
])
# Another way to define your optimizer
# 優(yōu)化函數(shù)
# 優(yōu)化算法用的是RMSprop
rmsprop = RMSprop(lr=0.001, rho=0.9, epsilon=1e-08, decay=0.0)
# We add metrics to get more results you want to see
# 不自己定義，直接用內(nèi)置的優(yōu)化器也行，optimizer='rmsprop'
#激活模型：接下來用 model.compile 激勵神經(jīng)網(wǎng)絡
model.compile(
optimizer=rmsprop,
loss='categorical_crossentropy',
metrics=['accuracy']
)
print('Training------------')
# Another way to train the model
# 訓練模型
# 上一個程序是用train_on_batch 一批一批的訓練 X_train, Y_train
# 默認的返回值是 cost，每100步輸出一下結果
# 輸出的樣式與上一個程序的有所不同，感覺用model.fit()更清晰明了
# 上一個程序是Python實現(xiàn)Keras搭建神經(jīng)網(wǎng)絡訓練回歸模型：
# https://blog.csdn.net/weixin_45798684/article/details/106503685
model.fit(X_train, y_train, nb_epoch=2, batch_size=32)
print('\nTesting------------')
# Evaluate the model with the metrics we defined earlier
# 測試
loss, accuracy = model.evaluate(X_test, y_test)
print('test loss:', loss)
print('test accuracy:', accuracy)
運行結果：
Using TensorFlow backend. Training------------ Epoch 1/2 32/60000 [..............................] - ETA: 5:03 - loss: 2.4464 - accuracy: 0.0625 864/60000 [..............................] - ETA: 14s - loss: 1.8023 - accuracy: 0.4850 1696/60000 [..............................] - ETA: 9s - loss: 1.5119 - accuracy: 0.6002 2432/60000 [>.............................] - ETA: 7s - loss: 1.3151 - accuracy: 0.6637 3200/60000 [>.............................] - ETA: 6s - loss: 1.1663 - accuracy: 0.7056 3968/60000 [>.............................] - ETA: 5s - loss: 1.0533 - accuracy: 0.7344 4704/60000 [=>............................] - ETA: 5s - loss: 0.9696 - accuracy: 0.7564 5408/60000 [=>............................] - ETA: 5s - loss: 0.9162 - accuracy: 0.7681 6112/60000 [==>...........................] - ETA: 5s - loss: 0.8692 - accuracy: 0.7804 6784/60000 [==>...........................] - ETA: 4s - loss: 0.8225 - accuracy: 0.7933 7424/60000 [==>...........................] - ETA: 4s - loss: 0.7871 - accuracy: 0.8021 8128/60000 [===>..........................] - ETA: 4s - loss: 0.7546 - accuracy: 0.8099 8960/60000 [===>..........................] - ETA: 4s - loss: 0.7196 - accuracy: 0.8183 9568/60000 [===>..........................] - ETA: 4s - loss: 0.6987 - accuracy: 0.8230 10144/60000 [====>.........................] - ETA: 4s - loss: 0.6812 - accuracy: 0.8262 10784/60000 [====>.........................] - ETA: 4s - loss: 0.6640 - accuracy: 0.8297 11456/60000 [====>.........................] - ETA: 4s - loss: 0.6462 - accuracy: 0.8329 12128/60000 [=====>........................] - ETA: 4s - loss: 0.6297 - accuracy: 0.8366 12704/60000 [=====>........................] - ETA: 4s - loss: 0.6156 - accuracy: 0.8405 13408/60000 [=====>........................] - ETA: 3s - loss: 0.6009 - accuracy: 0.8430 14112/60000 [======>.......................] - ETA: 3s - loss: 0.5888 - accuracy: 0.8457 14816/60000 [======>.......................] - ETA: 3s - loss: 0.5772 - accuracy: 0.8487 15488/60000 [======>.......................] - ETA: 3s - loss: 0.5685 - accuracy: 0.8503 16192/60000 [=======>......................] - ETA: 3s - loss: 0.5576 - accuracy: 0.8534 16896/60000 [=======>......................] - ETA: 3s - loss: 0.5477 - accuracy: 0.8555 17600/60000 [=======>......................] - ETA: 3s - loss: 0.5380 - accuracy: 0.8576 18240/60000 [========>.....................] - ETA: 3s - loss: 0.5279 - accuracy: 0.8600 18976/60000 [========>.....................] - ETA: 3s - loss: 0.5208 - accuracy: 0.8617 19712/60000 [========>.....................] - ETA: 3s - loss: 0.5125 - accuracy: 0.8634 20416/60000 [=========>....................] - ETA: 3s - loss: 0.5046 - accuracy: 0.8654 21088/60000 [=========>....................] - ETA: 3s - loss: 0.4992 - accuracy: 0.8669 21792/60000 [=========>....................] - ETA: 3s - loss: 0.4932 - accuracy: 0.8684 22432/60000 [==========>...................] - ETA: 3s - loss: 0.4893 - accuracy: 0.8693 23072/60000 [==========>...................] - ETA: 2s - loss: 0.4845 - accuracy: 0.8703 23648/60000 [==========>...................] - ETA: 2s - loss: 0.4800 - accuracy: 0.8712 24096/60000 [===========>..................] - ETA: 2s - loss: 0.4776 - accuracy: 0.8718 24576/60000 [===========>..................] - ETA: 2s - loss: 0.4733 - accuracy: 0.8728 25056/60000 [===========>..................] - ETA: 2s - loss: 0.4696 - accuracy: 0.8736 25568/60000 [===========>..................] - ETA: 2s - loss: 0.4658 - accuracy: 0.8745 26080/60000 [============>.................] - ETA: 2s - loss: 0.4623 - accuracy: 0.8753 26592/60000 [============>.................] - ETA: 2s - loss: 0.4600 - accuracy: 0.8756 27072/60000 [============>.................] - ETA: 2s - loss: 0.4566 - accuracy: 0.8763 27584/60000 [============>.................] - ETA: 2s - loss: 0.4532 - accuracy: 0.8771 28032/60000 [=============>................] - ETA: 2s - loss: 0.4513 - accuracy: 0.8775 28512/60000 [=============>................] - ETA: 2s - loss: 0.4477 - accuracy: 0.8784 28992/60000 [=============>................] - ETA: 2s - loss: 0.4464 - accuracy: 0.8786 29472/60000 [=============>................] - ETA: 2s - loss: 0.4439 - accuracy: 0.8791 29952/60000 [=============>................] - ETA: 2s - loss: 0.4404 - accuracy: 0.8800 30464/60000 [==============>...............] - ETA: 2s - loss: 0.4375 - accuracy: 0.8807 30784/60000 [==============>...............] - ETA: 2s - loss: 0.4349 - accuracy: 0.8813 31296/60000 [==============>...............] - ETA: 2s - loss: 0.4321 - accuracy: 0.8820 31808/60000 [==============>...............] - ETA: 2s - loss: 0.4301 - accuracy: 0.8827 32256/60000 [===============>..............] - ETA: 2s - loss: 0.4279 - accuracy: 0.8832 32736/60000 [===============>..............] - ETA: 2s - loss: 0.4258 - accuracy: 0.8838 33280/60000 [===============>..............] - ETA: 2s - loss: 0.4228 - accuracy: 0.8844 33920/60000 [===============>..............] - ETA: 2s - loss: 0.4195 - accuracy: 0.8849 34560/60000 [================>.............] - ETA: 2s - loss: 0.4179 - accuracy: 0.8852 35104/60000 [================>.............] - ETA: 2s - loss: 0.4165 - accuracy: 0.8854 35680/60000 [================>.............] - ETA: 2s - loss: 0.4139 - accuracy: 0.8860 36288/60000 [=================>............] - ETA: 2s - loss: 0.4111 - accuracy: 0.8870 36928/60000 [=================>............] - ETA: 2s - loss: 0.4088 - accuracy: 0.8874 37504/60000 [=================>............] - ETA: 2s - loss: 0.4070 - accuracy: 0.8878 38048/60000 [==================>...........] - ETA: 1s - loss: 0.4052 - accuracy: 0.8882 38656/60000 [==================>...........] - ETA: 1s - loss: 0.4031 - accuracy: 0.8888 39264/60000 [==================>...........] - ETA: 1s - loss: 0.4007 - accuracy: 0.8894 39840/60000 [==================>...........] - ETA: 1s - loss: 0.3997 - accuracy: 0.8896 40416/60000 [===================>..........] - ETA: 1s - loss: 0.3978 - accuracy: 0.8901 40960/60000 [===================>..........] - ETA: 1s - loss: 0.3958 - accuracy: 0.8906 41504/60000 [===================>..........] - ETA: 1s - loss: 0.3942 - accuracy: 0.8911 42016/60000 [====================>.........] - ETA: 1s - loss: 0.3928 - accuracy: 0.8915 42592/60000 [====================>.........] - ETA: 1s - loss: 0.3908 - accuracy: 0.8920 43168/60000 [====================>.........] - ETA: 1s - loss: 0.3889 - accuracy: 0.8924 43744/60000 [====================>.........] - ETA: 1s - loss: 0.3868 - accuracy: 0.8931 44288/60000 [=====================>........] - ETA: 1s - loss: 0.3864 - accuracy: 0.8931 44832/60000 [=====================>........] - ETA: 1s - loss: 0.3842 - accuracy: 0.8938 45408/60000 [=====================>........] - ETA: 1s - loss: 0.3822 - accuracy: 0.8944 45984/60000 [=====================>........] - ETA: 1s - loss: 0.3804 - accuracy: 0.8949 46560/60000 [======================>.......] - ETA: 1s - loss: 0.3786 - accuracy: 0.8953 47168/60000 [======================>.......] - ETA: 1s - loss: 0.3767 - accuracy: 0.8958 47808/60000 [======================>.......] - ETA: 1s - loss: 0.3744 - accuracy: 0.8963 48416/60000 [=======================>......] - ETA: 1s - loss: 0.3732 - accuracy: 0.8966 48928/60000 [=======================>......] - ETA: 0s - loss: 0.3714 - accuracy: 0.8971 49440/60000 [=======================>......] - ETA: 0s - loss: 0.3701 - accuracy: 0.8974 50048/60000 [========================>.....] - ETA: 0s - loss: 0.3678 - accuracy: 0.8979 50688/60000 [========================>.....] - ETA: 0s - loss: 0.3669 - accuracy: 0.8983 51264/60000 [========================>.....] - ETA: 0s - loss: 0.3654 - accuracy: 0.8988 51872/60000 [========================>.....] - ETA: 0s - loss: 0.3636 - accuracy: 0.8992 52608/60000 [=========================>....] - ETA: 0s - loss: 0.3618 - accuracy: 0.8997 53376/60000 [=========================>....] - ETA: 0s - loss: 0.3599 - accuracy: 0.9003 54048/60000 [==========================>...] - ETA: 0s - loss: 0.3583 - accuracy: 0.9006 54560/60000 [==========================>...] - ETA: 0s - loss: 0.3568 - accuracy: 0.9010 55296/60000 [==========================>...] - ETA: 0s - loss: 0.3548 - accuracy: 0.9016 56064/60000 [===========================>..] - ETA: 0s - loss: 0.3526 - accuracy: 0.9021 56736/60000 [===========================>..] - ETA: 0s - loss: 0.3514 - accuracy: 0.9026 57376/60000 [===========================>..] - ETA: 0s - loss: 0.3499 - accuracy: 0.9029 58112/60000 [============================>.] - ETA: 0s - loss: 0.3482 - accuracy: 0.9033 58880/60000 [============================>.] - ETA: 0s - loss: 0.3459 - accuracy: 0.9039 59584/60000 [============================>.] - ETA: 0s - loss: 0.3444 - accuracy: 0.9043 60000/60000 [==============================] - 5s 87us/step - loss: 0.3435 - accuracy: 0.9046 Epoch 2/2 32/60000 [..............................] - ETA: 11s - loss: 0.0655 - accuracy: 1.0000 736/60000 [..............................] - ETA: 4s - loss: 0.2135 - accuracy: 0.9389 1408/60000 [..............................] - ETA: 4s - loss: 0.2217 - accuracy: 0.9361 1984/60000 [..............................] - ETA: 4s - loss: 0.2316 - accuracy: 0.9390 2432/60000 [>.............................] - ETA: 4s - loss: 0.2280 - accuracy: 0.9379 3040/60000 [>.............................] - ETA: 4s - loss: 0.2374 - accuracy: 0.9368 3808/60000 [>.............................] - ETA: 4s - loss: 0.2251 - accuracy: 0.9386 4576/60000 [=>............................] - ETA: 4s - loss: 0.2225 - accuracy: 0.9379 5216/60000 [=>............................] - ETA: 4s - loss: 0.2208 - accuracy: 0.9377 5920/60000 [=>............................] - ETA: 4s - loss: 0.2173 - accuracy: 0.9383 6656/60000 [==>...........................] - ETA: 4s - loss: 0.2217 - accuracy: 0.9370 7392/60000 [==>...........................] - ETA: 4s - loss: 0.2224 - accuracy: 0.9360 8096/60000 [===>..........................] - ETA: 4s - loss: 0.2234 - accuracy: 0.9363 8800/60000 [===>..........................] - ETA: 3s - loss: 0.2235 - accuracy: 0.9358 9408/60000 [===>..........................] - ETA: 3s - loss: 0.2196 - accuracy: 0.9365 10016/60000 [====>.........................] - ETA: 3s - loss: 0.2207 - accuracy: 0.9363 10592/60000 [====>.........................] - ETA: 3s - loss: 0.2183 - accuracy: 0.9369 11168/60000 [====>.........................] - ETA: 3s - loss: 0.2177 - accuracy: 0.9377 11776/60000 [====>.........................] - ETA: 3s - loss: 0.2154 - accuracy: 0.9385 12544/60000 [=====>........................] - ETA: 3s - loss: 0.2152 - accuracy: 0.9393 13216/60000 [=====>........................] - ETA: 3s - loss: 0.2163 - accuracy: 0.9390 13920/60000 [=====>........................] - ETA: 3s - loss: 0.2155 - accuracy: 0.9391 14624/60000 [======>.......................] - ETA: 3s - loss: 0.2150 - accuracy: 0.9391 15424/60000 [======>.......................] - ETA: 3s - loss: 0.2143 - accuracy: 0.9398 16032/60000 [=======>......................] - ETA: 3s - loss: 0.2122 - accuracy: 0.9405 16672/60000 [=======>......................] - ETA: 3s - loss: 0.2096 - accuracy: 0.9409 17344/60000 [=======>......................] - ETA: 3s - loss: 0.2091 - accuracy: 0.9411 18112/60000 [========>.....................] - ETA: 3s - loss: 0.2086 - accuracy: 0.9416 18784/60000 [========>.....................] - ETA: 3s - loss: 0.2084 - accuracy: 0.9418 19392/60000 [========>.....................] - ETA: 3s - loss: 0.2076 - accuracy: 0.9418 20000/60000 [=========>....................] - ETA: 3s - loss: 0.2067 - accuracy: 0.9421 20608/60000 [=========>....................] - ETA: 3s - loss: 0.2071 - accuracy: 0.9419 21184/60000 [=========>....................] - ETA: 3s - loss: 0.2056 - accuracy: 0.9423 21856/60000 [=========>....................] - ETA: 3s - loss: 0.2063 - accuracy: 0.9419 22624/60000 [==========>...................] - ETA: 2s - loss: 0.2059 - accuracy: 0.9421 23328/60000 [==========>...................] - ETA: 2s - loss: 0.2056 - accuracy: 0.9422 23936/60000 [==========>...................] - ETA: 2s - loss: 0.2051 - accuracy: 0.9423 24512/60000 [===========>..................] - ETA: 2s - loss: 0.2041 - accuracy: 0.9424 25248/60000 [===========>..................] - ETA: 2s - loss: 0.2036 - accuracy: 0.9426 26016/60000 [============>.................] - ETA: 2s - loss: 0.2031 - accuracy: 0.9424 26656/60000 [============>.................] - ETA: 2s - loss: 0.2035 - accuracy: 0.9422 27360/60000 [============>.................] - ETA: 2s - loss: 0.2050 - accuracy: 0.9417 28128/60000 [=============>................] - ETA: 2s - loss: 0.2045 - accuracy: 0.9418 28896/60000 [=============>................] - ETA: 2s - loss: 0.2046 - accuracy: 0.9418 29536/60000 [=============>................] - ETA: 2s - loss: 0.2052 - accuracy: 0.9417 30208/60000 [==============>...............] - ETA: 2s - loss: 0.2050 - accuracy: 0.9417 30848/60000 [==============>...............] - ETA: 2s - loss: 0.2046 - accuracy: 0.9419 31552/60000 [==============>...............] - ETA: 2s - loss: 0.2037 - accuracy: 0.9421 32224/60000 [===============>..............] - ETA: 2s - loss: 0.2043 - accuracy: 0.9420 32928/60000 [===============>..............] - ETA: 2s - loss: 0.2041 - accuracy: 0.9420 33632/60000 [===============>..............] - ETA: 2s - loss: 0.2035 - accuracy: 0.9422 34272/60000 [================>.............] - ETA: 1s - loss: 0.2029 - accuracy: 0.9423 34944/60000 [================>.............] - ETA: 1s - loss: 0.2030 - accuracy: 0.9423 35648/60000 [================>.............] - ETA: 1s - loss: 0.2027 - accuracy: 0.9422 36384/60000 [=================>............] - ETA: 1s - loss: 0.2027 - accuracy: 0.9421 37120/60000 [=================>............] - ETA: 1s - loss: 0.2024 - accuracy: 0.9421 37760/60000 [=================>............] - ETA: 1s - loss: 0.2013 - accuracy: 0.9424 38464/60000 [==================>...........] - ETA: 1s - loss: 0.2011 - accuracy: 0.9424 39200/60000 [==================>...........] - ETA: 1s - loss: 0.2000 - accuracy: 0.9426 40000/60000 [===================>..........] - ETA: 1s - loss: 0.1990 - accuracy: 0.9428 40672/60000 [===================>..........] - ETA: 1s - loss: 0.1986 - accuracy: 0.9430 41344/60000 [===================>..........] - ETA: 1s - loss: 0.1982 - accuracy: 0.9432 42112/60000 [====================>.........] - ETA: 1s - loss: 0.1981 - accuracy: 0.9432 42848/60000 [====================>.........] - ETA: 1s - loss: 0.1977 - accuracy: 0.9433 43552/60000 [====================>.........] - ETA: 1s - loss: 0.1970 - accuracy: 0.9435 44256/60000 [=====================>........] - ETA: 1s - loss: 0.1972 - accuracy: 0.9436 44992/60000 [=====================>........] - ETA: 1s - loss: 0.1972 - accuracy: 0.9437 45664/60000 [=====================>........] - ETA: 1s - loss: 0.1966 - accuracy: 0.9438 46176/60000 [======================>.......] - ETA: 1s - loss: 0.1968 - accuracy: 0.9437 46752/60000 [======================>.......] - ETA: 1s - loss: 0.1969 - accuracy: 0.9438 47488/60000 [======================>.......] - ETA: 0s - loss: 0.1965 - accuracy: 0.9439 48256/60000 [=======================>......] - ETA: 0s - loss: 0.1965 - accuracy: 0.9438 48896/60000 [=======================>......] - ETA: 0s - loss: 0.1963 - accuracy: 0.9436 49568/60000 [=======================>......] - ETA: 0s - loss: 0.1962 - accuracy: 0.9438 50304/60000 [========================>.....] - ETA: 0s - loss: 0.1965 - accuracy: 0.9437 51072/60000 [========================>.....] - ETA: 0s - loss: 0.1967 - accuracy: 0.9437 51744/60000 [========================>.....] - ETA: 0s - loss: 0.1961 - accuracy: 0.9439 52480/60000 [=========================>....] - ETA: 0s - loss: 0.1957 - accuracy: 0.9439 53248/60000 [=========================>....] - ETA: 0s - loss: 0.1959 - accuracy: 0.9438 54016/60000 [==========================>...] - ETA: 0s - loss: 0.1963 - accuracy: 0.9437 54592/60000 [==========================>...] - ETA: 0s - loss: 0.1965 - accuracy: 0.9436 55168/60000 [==========================>...] - ETA: 0s - loss: 0.1962 - accuracy: 0.9436 55776/60000 [==========================>...] - ETA: 0s - loss: 0.1959 - accuracy: 0.9437 56448/60000 [===========================>..] - ETA: 0s - loss: 0.1965 - accuracy: 0.9437 57152/60000 [===========================>..] - ETA: 0s - loss: 0.1958 - accuracy: 0.9439 57824/60000 [===========================>..] - ETA: 0s - loss: 0.1956 - accuracy: 0.9438 58560/60000 [============================>.] - ETA: 0s - loss: 0.1951 - accuracy: 0.9440 59360/60000 [============================>.] - ETA: 0s - loss: 0.1947 - accuracy: 0.9440 60000/60000 [==============================] - 5s 76us/step - loss: 0.1946 - accuracy: 0.9440 Testing------------ 32/10000 [..............................] - ETA: 15s 1248/10000 [==>...........................] - ETA: 0s 2656/10000 [======>.......................] - ETA: 0s 4064/10000 [===========>..................] - ETA: 0s 5216/10000 [==============>...............] - ETA: 0s 6464/10000 [==================>...........] - ETA: 0s 7744/10000 [======================>.......] - ETA: 0s 9056/10000 [==========================>...] - ETA: 0s 9984/10000 [============================>.] - ETA: 0s 10000/10000 [==============================] - 0s 47us/step test loss: 0.17407772153392434 test accuracy: 0.9513000249862671
補充知識：Keras 搭建簡單神經(jīng)網(wǎng)絡：順序模型+回歸問題
多層全連接神經(jīng)網(wǎng)絡
每層神經(jīng)元個數(shù)、神經(jīng)網(wǎng)絡層數(shù)、激活函數(shù)等可自由修改
使用不同的損失函數(shù)可適用于其他任務，比如：分類問題
這是Keras搭建神經(jīng)網(wǎng)絡模型最基礎的方法之一，Keras還有其他進階的方法，官網(wǎng)給出了一些基本使用方法：Keras官網(wǎng)
# 這里搭建了一個4層全連接神經(jīng)網(wǎng)絡（不算輸入層），傳入函數(shù)以及函數(shù)內(nèi)部的參數(shù)均可自由修改
def ann(X, y):
'''
X: 輸入的訓練集數(shù)據(jù)
y: 訓練集對應的標簽
'''
'''初始化模型'''
# 首先定義了一個順序模型作為框架，然后往這個框架里面添加網(wǎng)絡層
# 這是最基礎搭建神經(jīng)網(wǎng)絡的方法之一
model = Sequential()
'''開始添加網(wǎng)絡層'''
# Dense表示全連接層，第一層需要我們提供輸入的維度 input_shape
# Activation表示每層的激活函數(shù)，可以傳入預定義的激活函數(shù)，也可以傳入符合接口規(guī)則的其他高級激活函數(shù)
model.add(Dense(64, input_shape=(X.shape[1],)))
model.add(Activation('sigmoid'))
model.add(Dense(256))
model.add(Activation('relu'))
model.add(Dense(256))
model.add(Activation('tanh'))
model.add(Dense(32))
model.add(Activation('tanh'))
# 輸出層，輸出的維度大小由具體任務而定
# 這里是一維輸出的回歸問題
model.add(Dense(1))
model.add(Activation('linear'))
'''模型編譯'''
# optimizer表示優(yōu)化器（可自由選擇），loss表示使用哪一種
model.compile(optimizer='rmsprop', loss='mean_squared_error')
# 自定義學習率，也可以使用原始的基礎學習率
reduce_lr = ReduceLROnPlateau(monitor='loss', factor=0.1, patience=10,
verbose=0, mode='auto', min_delta=0.001,
cooldown=0, min_lr=0)
'''模型訓練'''
# 這里的模型也可以先從函數(shù)返回后，再進行訓練
# epochs表示訓練的輪數(shù)，batch_size表示每次訓練的樣本數(shù)量（小批量學習），validation_split表示用作驗證集的訓練數(shù)據(jù)的比例
# callbacks表示回調(diào)函數(shù)的集合，用于模型訓練時查看模型的內(nèi)在狀態(tài)和統(tǒng)計數(shù)據(jù)，相應的回調(diào)函數(shù)方法會在各自的階段被調(diào)用
# verbose表示輸出的詳細程度，值越大輸出越詳細
model.fit(X, y, epochs=100,
batch_size=50, validation_split=0.0,
callbacks=[reduce_lr],
verbose=0)
# 打印模型結構
print(model.summary())
return model
下圖是此模型的結構圖，其中下劃線后面的數(shù)字是根據(jù)調(diào)用次數(shù)而定
以上這篇Python實現(xiàn)Keras搭建神經(jīng)網(wǎng)絡訓練分類模型教程就是小編分享給大家的全部內(nèi)容了，希望能給大家一個參考，也希望大家多多支持腳本之家。
相關文章
這篇文章主要介紹了Python設計模式之策略模式,結合實例形式分析了策略模式的概念、原理并結合實例形式分析了Python定義與使用策略模式相關操作技巧,需要的朋友可以參考下2019-01-01
python使用in操作符時元組和數(shù)組的區(qū)別分析
有時候要判斷一個數(shù)是否在一個序列里面，這時就會用到in運算符來判斷成員資格，如果條件為真時，就會返回true，條件為假時，返回一個flase。這樣的運算符叫做布爾運算符，其真值叫做布爾值。2015-05-05
這篇文章主要介紹了詳解PyQt5信號與槽的幾種高級玩法，文中通過示例代碼介紹的非常詳細，對大家的學習或者工作具有一定的參考學習價值，需要的朋友們下面隨著小編來一起學習學習吧2020-03-03
用Python從零實現(xiàn)貝葉斯分類器的機器學習的教程
這篇文章主要介紹了用Python從零實現(xiàn)貝葉斯分類器的教程,樸素貝葉斯算法屬于機器學習中的基礎內(nèi)容、實用而高效，本文詳細展示了用Python語言實現(xiàn)的步驟，需要的朋友可以參考下2015-03-03
今天小編就為大家分享一篇Python: 傳遞列表副本方式，具有很好的參考價值，希望對大家有所幫助。一起跟隨小編過來看看吧2019-12-12
這篇文章主要介紹了Flask URL傳參與視圖映射的實現(xiàn)方法，文中通過示例代碼介紹的非常詳細，對大家的學習或者工作具有一定的參考學習價值，需要的朋友們下面隨著小編來一起學習吧2023-03-03
接下來，由于我覺得手動復制粘貼這29000條插入語句實在是太痛苦了，所以打算用Python來完成這項工作。這是我第一次自己動手寫Python代碼，感覺還挺順利的。2008-04-04
python如何將兩個數(shù)據(jù)表中的對應數(shù)據(jù)相加
這篇文章主要介紹了python如何將兩個數(shù)據(jù)表中的對應數(shù)據(jù)相加問題,具有很好的參考價值,希望對大家有所幫助,如有錯誤或未考慮完全的地方,望不吝賜教2023-08-08
conda創(chuàng)建pytorch環(huán)境報錯
這篇文章主要介紹了conda創(chuàng)建pytorch環(huán)境報錯，幫助大家更好的理解和學習使用python，感興趣的朋友可以了解下2021-04-04
這篇文章主要為大家分享了python web框架學習筆記，感興趣的小伙伴們可以參考一下2016-05-05