keras训练网络回归角度，仅作个人笔记。

最新推荐文章于 2023-04-21 11:41:10 发布
最新推荐文章于 2023-04-21 11:41:10 发布
阅读量2.4k 收藏
点赞数
版权声明：本文为博主原创文章，遵循 CC 4.0 BY-SA 版权协议，转载请附上原文出处链接和本声明。
本文链接：https://blog.csdn.net/qq_36289191/article/details/94768893
版权 
import numpy as np
import scipy
import os
import keras
from keras.models import Sequential
from keras.layers import Dense, Dropout, Flatten
from keras.layers import Conv2D, MaxPooling2D,AveragePooling2D
from keras.layers import BatchNormalization,Activation
from keras.optimizers import SGD
from keras.callbacks import ReduceLROnPlateau, ModelCheckpoint
from keras import backend as K
from keras.utils import np_utils
from matplotlib import pyplot as plt
from PIL import Image
n = 10000
index=[i for i in range(10000)]
#print(index)
#print("***************")
np.random.shuffle(index)
fileresult=open('photo/label.txt','r')
array=fileresult.readlines()
def data_generator(batch_size):
    '''data generator for fit_generator'''
    # n = 5000
    # index=[i for i in range(5000)]
    # print(index)
    # print("***************")
    # np.random.shuffle(index)
    # #i = 0
    # print(index)
    
    k=0
    # fileresult=open('pho/label.txt','r')
    # array=fileresult.readlines()
    # print(len(array))
    while True:
        x= []
        y = []

        #print("a")
        for b in range(batch_size):
            #print("b")
            #index=np.random.randint(0, n-1)
            c=b+batch_size*k
            #print(k,b,c,index[c])
            file_path='photo/{}.png'.format(index[c]+1)
            img=Image.open(file_path)
            #print(index[c]+1)
            #print("c")
            #img=img.resize((224, 224),Image.ANTIALIAS)
            #ng = img.transpose(Image.FLIP_LEFT_RIGHT)
            #print("c:",c,"k:",k)
            arr=np.array(img)
            arr=arr/255.0
            x.append(arr)
            #arr=np.array(ng)
            #arr=arr/255.0
            #x.append(arr)
            #print("d")
            c_array=array[c].split(" ")
            #print("e")
            #print(c_array)
            tt=float(c_array[-1][0:-1])
            #print(tt)
            y.append(tt)
            #y.append((180-tt if tt>0 else -180-tt)/180)
            #i = (i+1) % n
        x = np.array(x)
        #print(x)
        y = np.array(y)
        #print(y)
        x=np.expand_dims(x, axis=3);
        y=np.expand_dims(y, axis=1);
        k+=1
        if(k==10000//128):
            k=0
        yield [x, y]
def feature_normalize(dataset):



    mu = np.mean(dataset)

    sigma = np.std(dataset)

    return (dataset - mu)/sigma
cwd='photo/'
x_train=[]
for id in range(1,5001):
    file_path='photo/{}.png'.format(id)
    line=[]
    img=Image.open(file_path)
        #print(img.size)
        #print(img.mode)
    arr=np.array(img)
    arr=arr/255.0
        # plt.figure("box")
        # plt.imshow(img)
        # plt.show()
    x_train.append(arr)
        #np.array(x_train)

x_test=[]
for id in range(5001,10001):
        file_path='photo/{}.txt'.format(id)
        line=[]
        fileresult=open(file_path,'r')
        for c in fileresult.readlines():
            lc=c[1:-2].split(' ')
            #print(lc)
            tt=map(float,lc)
            line.append(tt)
        x_test.append(line)
y_train=[]
fileresult=open('photo/result.txt','r')
for c in fileresult.readlines():
    tt=int(c)
    y_train.append(tt)
y_test=[]
fileresult=open('photo/result2.txt','r')
for c in fileresult.readlines():
    tt=int(c)
    y_test.append(tt)
#print(y_train)
x_train=np.expand_dims(x_train, axis=3);
y_train=np.expand_dims(y_train, axis=1);
x_test=np.expand_dims(x_test, axis=3);
y_test=np.expand_dims(y_test, axis=1);
#print(x_train.shape)
#print(y_train)
model = Sequential()
model.add(Conv2D(8, (3, 3), padding='same',
                 input_shape=x_test.shape[1:]))
model.add(BatchNormalization(axis = 3))
model.add(Activation('relu'))
model.add(AveragePooling2D(pool_size=(2, 2),strides=2))
model.add(Conv2D(16, (3, 3),padding='same'))
model.add(BatchNormalization(axis = 3))
model.add(Activation('relu'))
model.add(AveragePooling2D(pool_size=(2, 2),strides=2))
model.add(Conv2D(32, (3, 3),padding='same'))
model.add(BatchNormalization(axis = 3))
model.add(Activation('relu'))
model.add(Conv2D(32, (3, 3),padding='same'))
model.add(BatchNormalization(axis = 3))
model.add(Activation('relu'))
model.add(Dropout(0.2))
model.add(Flatten())
model.add(Dense(1))
sgd = SGD(lr=0.000001, decay=1e-7, momentum=0.9, nesterov=True)
model.compile(loss='mean_squared_error', optimizer=sgd)
print(model.summary())
callbacks_list = [

    keras.callbacks.ModelCheckpoint(

        filepath='best_model.{epoch:02d}-{val_loss:.2f}.h5',

        monitor='val_loss', save_best_only=True),

    keras.callbacks.EarlyStopping(monitor='val_loss', patience=10)

]
BATCH_SIZE = 8

EPOCHS = 300
print("shapes\n")


validation_split=0.1
num_val=10000*validation_split
num_train=10000-num_val

#print(x_train.shape)
#print(y_train.shape)
# history = model.fit_generator(data_generator(BATCH_SIZE),

#                             steps_per_epoch=max(1, num_train//BATCH_SIZE),

#                             validation_data=data_generator(BATCH_SIZE),

#                             validation_steps=max(1, num_val//BATCH_SIZE ),
#                             epochs=EPOCHS,
#                             initial_epoch=0,

#                             callbacks=callbacks_list,


#                             )
#print(x_train.shape)
#print(y_train.shape)
history = model.fit(x_train,

                      y_train,

                      batch_size=BATCH_SIZE,

                      epochs=EPOCHS,

                      callbacks=callbacks_list,
                      shuffle=True,

                      validation_split=0.1,

                      verbose=1)
output = model.predict(x_test)
err=map(abs,output-y_test)
print(len(err))
num=0
num1=0
num2=0
for c in err:
    if c<20:
        num+=1
    if c<45:
        num1+=1
    if c<10:
        num2+=1
print(num/5000.0)
print(num1/5000.0)
print(num2/5000.0)
plt.figure(figsize=(6, 4))

#plt.plot(history.history['acc'], "g--", label="Accuracy of training data")

#plt.plot(history.history['val_acc'], "g", label="Accuracy of validation data")

plt.plot(history.history['loss'], "r--", label="Loss of training data")

plt.plot(history.history['val_loss'], "r", label="Loss of validation data")

plt.title('Model Accuracy and Loss')

plt.ylabel('Accuracy and Loss')

plt.xlabel('Training Epoch')

plt.ylim(0)

plt.legend()

plt.show()
#model.load_weights('best_model.30-22.85.h5', by_name=False)

训练参数：

_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_1 (Conv2D)            (None, 28, 28, 8)         80        
_________________________________________________________________
batch_normalization_1 (Batch (None, 28, 28, 8)         32        
_________________________________________________________________
activation_1 (Activation)    (None, 28, 28, 8)         0         
_________________________________________________________________
average_pooling2d_1 (Average (None, 14, 14, 8)         0         
_________________________________________________________________
conv2d_2 (Conv2D)            (None, 14, 14, 16)        1168      
_________________________________________________________________
batch_normalization_2 (Batch (None, 14, 14, 16)        64        
_________________________________________________________________
activation_2 (Activation)    (None, 14, 14, 16)        0         
_________________________________________________________________
average_pooling2d_2 (Average (None, 7, 7, 16)          0         
_________________________________________________________________
conv2d_3 (Conv2D)            (None, 7, 7, 32)          4640      
_________________________________________________________________
batch_normalization_3 (Batch (None, 7, 7, 32)          128       
_________________________________________________________________
activation_3 (Activation)    (None, 7, 7, 32)          0         
_________________________________________________________________
conv2d_4 (Conv2D)            (None, 7, 7, 32)          9248      
_________________________________________________________________
batch_normalization_4 (Batch (None, 7, 7, 32)          128       
_________________________________________________________________
activation_4 (Activation)    (None, 7, 7, 32)          0         
_________________________________________________________________
dropout_1 (Dropout)          (None, 7, 7, 32)          0         
_________________________________________________________________
flatten_1 (Flatten)          (None, 1568)              0         
_________________________________________________________________
dense_1 (Dense)              (None, 1)                 1569      
=================================================================
Total params: 17,057
Trainable params: 16,881
Non-trainable params: 176
_________________________________________________________________

训练结果：

Epoch 1/300
2019-07-05 21:36:38.061295: I tensorflow/stream_executor/dso_loader.cc:152] successfully opened CUDA library libcublas.so.10.0 locally
4500/4500 [==============================] - 4s 828us/step - loss: 529.2516 - val_loss: 412.2183
Epoch 2/300
4500/4500 [==============================] - 2s 499us/step - loss: 261.9935 - val_loss: 205.1285
Epoch 3/300
4500/4500 [==============================] - 2s 522us/step - loss: 183.9583 - val_loss: 176.4640
Epoch 4/300
4500/4500 [==============================] - 2s 532us/step - loss: 157.9902 - val_loss: 165.5905
Epoch 5/300
4500/4500 [==============================] - 3s 600us/step - loss: 142.2254 - val_loss: 133.4456
Epoch 6/300
4500/4500 [==============================] - 3s 574us/step - loss: 126.0722 - val_loss: 120.4842
Epoch 7/300
4500/4500 [==============================] - 3s 561us/step - loss: 115.6117 - val_loss: 113.8038
Epoch 8/300
4500/4500 [==============================] - 2s 524us/step - loss: 103.8097 - val_loss: 110.5109
Epoch 9/300
4500/4500 [==============================] - 2s 509us/step - loss: 99.2515 - val_loss: 102.6878
Epoch 10/300
4500/4500 [==============================] - 2s 493us/step - loss: 90.2994 - val_loss: 104.8946
Epoch 11/300
4500/4500 [==============================] - 3s 561us/step - loss: 84.4916 - val_loss: 84.4557
Epoch 12/300
4500/4500 [==============================] - 3s 593us/step - loss: 80.1074 - val_loss: 98.9638
Epoch 13/300
4500/4500 [==============================] - 2s 553us/step - loss: 75.9412 - val_loss: 88.8961
Epoch 14/300
4500/4500 [==============================] - 2s 496us/step - loss: 71.1955 - val_loss: 71.7264
Epoch 15/300
4500/4500 [==============================] - 2s 508us/step - loss: 67.3599 - val_loss: 65.1363
Epoch 16/300
4500/4500 [==============================] - 3s 556us/step - loss: 63.9384 - val_loss: 74.2647
Epoch 17/300
4500/4500 [==============================] - 2s 500us/step - loss: 60.4251 - val_loss: 64.9483
Epoch 18/300
4500/4500 [==============================] - 2s 492us/step - loss: 58.4589 - val_loss: 79.7609
Epoch 19/300
4500/4500 [==============================] - 2s 495us/step - loss: 56.2028 - val_loss: 57.1882
Epoch 20/300
4500/4500 [==============================] - 2s 488us/step - loss: 54.4542 - val_loss: 54.6668
Epoch 21/300
4500/4500 [==============================] - 2s 495us/step - loss: 52.0168 - val_loss: 65.2611
Epoch 22/300
4500/4500 [==============================] - 2s 493us/step - loss: 49.6020 - val_loss: 56.3058
Epoch 23/300
4500/4500 [==============================] - 2s 492us/step - loss: 48.6968 - val_loss: 61.4040
Epoch 24/300
4500/4500 [==============================] - 2s 484us/step - loss: 47.2439 - val_loss: 51.7113
Epoch 25/300
4500/4500 [==============================] - 2s 496us/step - loss: 45.8646 - val_loss: 51.0462
Epoch 26/300
4500/4500 [==============================] - 2s 496us/step - loss: 43.7266 - val_loss: 57.8190
Epoch 27/300
4500/4500 [==============================] - 2s 493us/step - loss: 41.8024 - val_loss: 52.0391
Epoch 28/300
4500/4500 [==============================] - 2s 486us/step - loss: 41.0792 - val_loss: 49.6890
Epoch 29/300
4500/4500 [==============================] - 2s 493us/step - loss: 39.9201 - val_loss: 45.2138
Epoch 30/300
4500/4500 [==============================] - 2s 492us/step - loss: 38.7481 - val_loss: 42.9585
Epoch 31/300
4500/4500 [==============================] - 2s 494us/step - loss: 39.3088 - val_loss: 51.3753
Epoch 32/300
4500/4500 [==============================] - 2s 497us/step - loss: 36.7096 - val_loss: 45.1476
Epoch 33/300
4500/4500 [==============================] - 2s 491us/step - loss: 36.1465 - val_loss: 42.1468
Epoch 34/300
4500/4500 [==============================] - 2s 491us/step - loss: 36.1388 - val_loss: 41.8209
Epoch 35/300
4500/4500 [==============================] - 2s 502us/step - loss: 35.4413 - val_loss: 41.0889
Epoch 36/300
4500/4500 [==============================] - 2s 490us/step - loss: 34.7038 - val_loss: 41.5128
Epoch 37/300
4500/4500 [==============================] - 2s 493us/step - loss: 33.0647 - val_loss: 42.6657
Epoch 38/300
4500/4500 [==============================] - 2s 495us/step - loss: 32.4053 - val_loss: 40.4518
Epoch 39/300
4500/4500 [==============================] - 2s 487us/step - loss: 31.6384 - val_loss: 45.8528
Epoch 40/300
4500/4500 [==============================] - 2s 493us/step - loss: 31.7787 - val_loss: 38.3601
Epoch 41/300
4500/4500 [==============================] - 2s 499us/step - loss: 31.7154 - val_loss: 39.3345
Epoch 42/300
4500/4500 [==============================] - 2s 486us/step - loss: 31.1642 - val_loss: 43.8633
Epoch 43/300
4500/4500 [==============================] - 2s 495us/step - loss: 29.6413 - val_loss: 37.3922
Epoch 44/300
4500/4500 [==============================] - 2s 490us/step - loss: 29.5115 - val_loss: 41.6811
Epoch 45/300
4500/4500 [==============================] - 2s 488us/step - loss: 28.8164 - val_loss: 46.3363
Epoch 46/300
4500/4500 [==============================] - 2s 494us/step - loss: 28.7848 - val_loss: 45.4383
Epoch 47/300
4500/4500 [==============================] - 2s 496us/step - loss: 27.6349 - val_loss: 41.1156
Epoch 48/300
4500/4500 [==============================] - 2s 485us/step - loss: 26.6820 - val_loss: 36.0883
Epoch 49/300
4500/4500 [==============================] - 2s 488us/step - loss: 26.6154 - val_loss: 34.9007
Epoch 50/300
4500/4500 [==============================] - 2s 494us/step - loss: 26.0024 - val_loss: 36.3475
Epoch 51/300
4500/4500 [==============================] - 2s 490us/step - loss: 26.3216 - val_loss: 35.7313
Epoch 52/300
4500/4500 [==============================] - 2s 493us/step - loss: 26.1783 - val_loss: 35.7962
Epoch 53/300
4500/4500 [==============================] - 2s 495us/step - loss: 26.0408 - val_loss: 33.8485
Epoch 54/300
4500/4500 [==============================] - 2s 494us/step - loss: 25.8208 - val_loss: 51.6296
Epoch 55/300
4500/4500 [==============================] - 2s 498us/step - loss: 24.6244 - val_loss: 35.7905
Epoch 56/300
4500/4500 [==============================] - 2s 496us/step - loss: 24.4153 - val_loss: 36.1860
Epoch 57/300
4500/4500 [==============================] - 2s 497us/step - loss: 24.0448 - val_loss: 42.8698
Epoch 58/300
4500/4500 [==============================] - 2s 498us/step - loss: 24.4918 - val_loss: 35.1314
Epoch 59/300
4500/4500 [==============================] - 2s 495us/step - loss: 22.7746 - val_loss: 33.3021
Epoch 60/300
4500/4500 [==============================] - 2s 488us/step - loss: 23.2578 - val_loss: 34.9876
Epoch 61/300
4500/4500 [==============================] - 2s 498us/step - loss: 22.7299 - val_loss: 33.7600
Epoch 62/300
4500/4500 [==============================] - 2s 501us/step - loss: 23.5896 - val_loss: 36.9000
Epoch 63/300
4500/4500 [==============================] - 2s 491us/step - loss: 22.5004 - val_loss: 34.7175
Epoch 64/300
4500/4500 [==============================] - 2s 518us/step - loss: 22.4151 - val_loss: 31.8431
Epoch 65/300
4500/4500 [==============================] - 2s 528us/step - loss: 22.5390 - val_loss: 32.5396
Epoch 66/300
4500/4500 [==============================] - 3s 558us/step - loss: 21.3262 - val_loss: 36.8819
Epoch 67/300
4500/4500 [==============================] - 2s 505us/step - loss: 21.7721 - val_loss: 35.8961
Epoch 68/300
4500/4500 [==============================] - 2s 490us/step - loss: 21.8546 - val_loss: 32.9064
Epoch 69/300
4500/4500 [==============================] - 2s 507us/step - loss: 21.8813 - val_loss: 36.9778
Epoch 70/300
4500/4500 [==============================] - 2s 495us/step - loss: 21.4432 - val_loss: 29.8957
Epoch 71/300
4500/4500 [==============================] - 2s 492us/step - loss: 21.2658 - val_loss: 35.6055
Epoch 72/300
4500/4500 [==============================] - 2s 486us/step - loss: 20.4156 - val_loss: 31.9169
Epoch 73/300
4500/4500 [==============================] - 2s 491us/step - loss: 20.3294 - val_loss: 34.3695
Epoch 74/300
4500/4500 [==============================] - 2s 489us/step - loss: 19.9095 - val_loss: 32.4377
Epoch 75/300
4500/4500 [==============================] - 2s 490us/step - loss: 20.2715 - val_loss: 31.1379
Epoch 76/300
4500/4500 [==============================] - 2s 495us/step - loss: 19.5925 - val_loss: 30.7104
Epoch 77/300
4500/4500 [==============================] - 2s 489us/step - loss: 19.2491 - val_loss: 33.6524
Epoch 78/300
4500/4500 [==============================] - 2s 495us/step - loss: 19.8310 - val_loss: 34.4298
Epoch 79/300
4500/4500 [==============================] - 2s 534us/step - loss: 19.2930 - val_loss: 35.9891
Epoch 80/300
4500/4500 [==============================] - 2s 526us/step - loss: 20.0375 - val_loss: 30.6407
5000
0.9974
1.0
0.9282

 

墨藍
关注
  • 0
    点赞
  • 踩
  • 0
    收藏
    觉得还不错? 一键收藏
  • 0
    评论
基于BP回归神经网络的人体角度拟合研究
05-06
基于深度学习和深度摄像机的人体动作识别方法，受其应用场景所限，均不能对视频中快变场景和静态图像中的人体动作进行识别.本文中定义了人体肢体角度空间，使用基于深度学习的人体骨骼识别框架的骨骼数据，构建8个4层BP回归神经网络.对人体的骨骼数据提取和预处理后，再对训练数据进行增维处理，通过回归神经网络进行拟合，实验和测试结果表明，该方法可以有效的对人体角度进行回归，为快变场景和静态图像中的人的动作识别提供可靠依据.
MATLAB 使用CNN拟合回归模型预测手写数字的旋转角度（卷积神经网络）
qingfengxd1的博客
05-09 5495
学习来源自mathworks的官方范例，个人学习使用，在个人项目上可以按照需求变化数据集来实现CNN回归计算 数据集生成方法可以参考：https://blog.csdn.net/qingfengxd1/article/details/105931988 %% 加载数据 %% 数据集包含手写数字的合成图像，以及每幅图像旋转的对应角度(以角度为单位)。 %% 使用digitTrain4DArrayData和digitTest4DArrayData将训练和验证图像加载为4D数组。 %% 输出YTrain和Y
tensorflow的Data Pipeline系列教程（一）——Dataset类的属性即常用方法
MIss-Y的博客
06-13 5286
前言：在tensorflow中，训练数据常常需要经过随机打乱、分成一个一个的batch来进行训练，当然有很多的方式可以完成，比如我们可以通过传统的python方法构建迭代器，我们也可以使用其它的一些方法，但是tensorflow本身提供了强大的data pipeline处理机制，本文就来详细说明一下。本文的核心是：从Dataset到Iterator的“四步走”过程。 ...
基于Keras框架的自定义数据集卷积神经网络(CNN)用于回归预测，并讨论调参减小模型大小送入stm32板子
qq_48411488的博客
04-21 1276
搭建可以送入stm32板子的keras框架的CNN网络，使用自定义数据集做回归预测，自己编写了完整的CNN代码。由于是新手刚学习python，有不对或者代码繁琐的请多多指教。
1.5 条件生成对抗网络（CGAN)
fdgdg1的博客
11-22 3408
1.什么是CGAN 在CGAN训练期间，生成器学习为训练数据集中的每个标签生成逼真的样本，而鉴别器则学习区分真的样本-标签对与假的样本-标签对。只学习接受真实且样本-标签匹配正确的对，拒绝不匹配的对和样本为假的对。 2.生成器 条件标签称为y，生成器使用噪声向量Z和标签y合成一个伪样本。 3.鉴别器 接受带标签的真实样本（x,y),以及生成器生成的伪样本 4.代码实现 4.1导入数据库 %matplotlib inline imp...
keras手势识别神经网络训练.zip
最新发布
03-06
用keras训练识别手势的神经网络。 可自定义录制手势
keras 神经网络解决回归问题实例_波士顿房价预测.rar
03-21
keras 神经网络解决回归问题实例_波士顿房价预测
Keras——用Keras搭建线性回归神经网络
12-21
用Keras搭建线性回归神经网络2.1.导入必要模块2.2.创建数据2.3.搭建模型2.4.激活模型2.5.训练+测试 1.前言 神经网络可以用来模拟回归问题 (regression)，例如给一组数据，用一条线来对数据进行拟合，并可以预测新...
keras训练浅层卷积网络并保存和加载模型实例
09-16
主要介绍了keras训练浅层卷积网络并保存和加载模型实例，具有很好的参考价值，希望对大家有所帮助。一起跟随小编过来看看吧
Keras 实现加载预训练模型并冻结网络的层
12-17
比如，分类任务中，优异的深度学习网络有很多。 ResNet, VGG, Xception等等… 并且这些模型参数已经在imagenet数据集中训练的很好了，可以直接拿过来用。 根据自己的任务，训练一下最后的分类层即可得到比较好的结果...
基于深度学习的点定位回归和角度预测的车位检测
06-27
基于深度学习的点定位回归和角度预测的车位检测
如何使用卷积神经网络拟合一个回归模型来预测手写数字的旋转角度
03-09
如何使用卷积神经网络拟合一个回归模型来预测手写数字的旋转角度。
Fashion数据集卷积报错：Input 0 of layer conv2d is incompatible with the layer: expected ndim=4, found ndim=3
林风风
05-04 2万+
做中国大学MOOC中“人工智能实践：TensorFlow笔记”的fashion数据集卷积神经网络练习时报如下错误： ValueError: Input 0 of layer conv2d is incompatible with the layer: expected ndim=4, found ndim=3. Full shape received: [None, 28, 28] 但是一模一...
使用keras做线性回归训练记录
tugouxp的专栏
09-25 240
我们建立一个如下的模型： 数学表达师式是 在这里，我们给定 并且给定一组按照此规律分布的数据，用这组数据进行训练，看出来的网络能否作出准确预测。 数据输入: [[-1.0,2.0],[0.0,1.0],[2.0,3.0],[4.0,0.9]] 数据输出：[15.0,9.0,29.0,16.2] 那么当数据是两组的[4.0,11.0],[5.9,6.7]时候，输出是什么呢？ 训练模型： from tensorflow import keras import numpy as n.
【入门教程】使用keras实现线性回归
qq_43018832的博客
09-29 1325
【代码】【入门教程】使用keras实现线性回归。
python神经网络模型调用后预测值不变_我的Keras使用总结（1）——Keras概述与常见问题整理,Python机器学习笔记：利用Keras进行分类预测,Python机器学习笔记：使用Keras进...
weixin_39932838的博客
12-05 1541
今天整理了自己所写的关于Keras的博客，有没发布的，有发布的，但是整体来说是有点乱的。上周有空，认真看了一周Keras的中文文档，稍有心得，整理于此。这里附上Keras官网地址：下面回顾一下自己以前写的有关Keras的博客：首先是上面两篇博客，这两篇写出来，其实是整理笔记，单纯的学习一下基础的回归和分类，使用Keras创建一个简单的回归问题的神经网络模型和多类分类的模型。使用了最经典的回归数据波...
5 python numpy.expand_dims的用法
热门推荐
hxshine的博客
12-01 10万+
1 查看help其实感觉expand_dims(a, axis)就是在axis的那一个轴上把数据加上去，这个数据在axis这个轴的0位置。 例如原本为一维的2个数据，axis=0，则shape变为(1,2),axis=1则shape变为(2,1) 再例如 原本为 (2,3),axis=0，则shape变为(1,2,3),axis=1则shape变为(2,1,3)help(np.expand_di
Keras搭建Regressor回归神经网络
慢慢慢慢曼
03-27 3743
写在前面： 刚刚开始学习keras，就在这里记录一下学习过程啦。 本文为使用keras搭建回归神经网络，对numpy的使用也有了一定的了解。 keras文档：https://keras.io/zh/ numpy文档：https://www.numpy.org.cn/index.html 莫烦python bilibili链接：https://space.bilibili.com/243821484...
用keras创建拟合网络解决回归问题Regression
csmqq的专栏
05-16 2万+
实现了正弦曲线的拟合，即regression问题。 创建的模型单输入单输出，两个隐层分别为100、50个神经元。 在keras的官方文档中，给的例子多是关于分类的。因此在测试regression时，遇到了一些问题。总结来说，应注意以下几个方面：
keras训练集制作
01-19
在使用Keras进行训练集制作的过程中，首先需要准备好原始数据集。这些数据可以是图像、文本、语音等各种形式。然后，需要对数据集进行处理，比如标准化、归一化、去噪等操作，以保证数据的质量和可用性。 接下来，...

“相关推荐”对你有帮助么？

  • 非常没帮助
  • 没帮助
  • 一般
  • 有帮助
  • 非常有帮助
提交
评论
添加红包

请填写红包祝福语或标题

个

红包个数最小为10个

元

红包金额最低5元

当前余额3.43元 前往充值 >
需支付：10.00元
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付元
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明：

1.余额是钱包充值的虚拟货币，按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载，可以购买VIP、付费专栏及课程。

余额充值