DeepHao

Tensorflow 笔记 Ⅵ——Titanic Keras建模与应用

文章目录

数据集
基本原理
Titanic TensorFLow 2.x Keras API 实现

前情函数介绍

Pandas 缺失值判断函数 isnull()、isnull().any()、isnull().sum()
Pandas 缺失值填充函数 fillna()
Pandas map() 映射函数
Pandas sample() 数据洗牌
Pandas drop() 删除数据
Pandas values 转换为 ndarray

正式开始

导入必要包
数据下载
读取数据
数据预处理

空值填充
数字映射

创建训练数据集与标签
数据标准化

代码重构定义数据预处理函数

数据读取并洗牌
训练可视化
测试模型
使用 Jack & Rose 测试
数据插入最后一列生存概率

加入回调

加载模型

Titanic TensorFLow 1.x Keras API 实现

导入必要的包
定义预处理函数
读取数据，制作数据集
搭建模型
开始训练
训练可视化
模型预测
加载模型进行预测

Tensorflow 笔记 Ⅵ——Titanic Keras建模与应用_第1张图片

数据集

数据集记录了泰坦尼克号一部分人的信息，以及其存活率等，众所周知，泰坦尼克号是一场海难，也就造成了人员信息难以调查，所以数据集中具有一些缺失的数据，数据集可以在 Kaggle 下载，也可以点击此处下载

	pclass	survived	name	sex	age	sibsp	parch	ticket	fare	cabin	embarked	boat	body	home.dest
0	1	1	Allen, Miss. Elisabeth Walton	female	29.0000	0	0	24160	211.3375	B5	S	2	NaN	St Louis, MO
1	1	1	Allison, Master. Hudson Trevor	male	0.9167	1	2	113781	151.5500	C22 C26	S	11	NaN	Montreal, PQ / Chesterville, ON
2	1	0	Allison, Miss. Helen Loraine	female	2.0000	1	2	113781	151.5500	C22 C26	S	NaN	NaN	Montreal, PQ / Chesterville, ON
3	1	0	Allison, Mr. Hudson Joshua Creighton	male	30.0000	1	2	113781	151.5500	C22 C26	S	NaN	135.0	Montreal, PQ / Chesterville, ON
4	1	0	Allison, Mrs. Hudson J C (Bessie Waldo Daniels)	female	25.0000	1	2	113781	151.5500	C22 C26	S	NaN	NaN	Montreal, PQ / Chesterville, ON
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
1304	3	0	Zabour, Miss. Hileni	female	14.5000	1	0	2665	14.4542	NaN	C	NaN	328.0	NaN
1305	3	0	Zabour, Miss. Thamine	female	NaN	1	0	2665	14.4542	NaN	C	NaN	NaN	NaN
1306	3	0	Zakarian, Mr. Mapriededer	male	26.5000	0	0	2656	7.2250	NaN	C	NaN	304.0	NaN
1307	3	0	Zakarian, Mr. Ortin	male	27.0000	0	0	2670	7.2250	NaN	C	NaN	NaN	NaN
1308	3	0	Zimmerman, Mr. Leo	male	29.0000	0	0	315082	7.8750	NaN	S	NaN	NaN	NaN

1309 rows × 14 columns

泰坦尼克标签说明

字段	字段说明	数据说明
pclass	舱等级	1头等舱、2二等舱、3三等舱
survival	是否生存	0否、1是
name	姓名	None
sex	性别	Female女性、male男性
age	年龄	None
sibsp	siblings + parents	兄弟姐妹或父母是否同船
parch	parents + children	parents父母、children孩子
ticked	船票号码	None
fare	船票费用	None
cabin	舱位号码	None
embarked	登船港口	C=Cherbourg，Q=Queenstown，S=Southampton
home.dest	家、目的地	home、destination

基本原理

利用了多元线性回归问题的预处理方式，在最后一层输出层划分为分类问题，有关多元线性回归与逻辑回归问题的原理，屈尊移驾这里
和那里，由于此问题属于逻辑回归中的二分类问题，所以激活函数选用了 Sigmod 函数，在上诉的“那里”连接中解释了 Sigmod 函数

模型采用全连接方式，两个隐藏层，一个神经元输出作为预测的概率

Titanic TensorFLow 2.x Keras API 实现

前情函数介绍

在 data 文件夹下，新建一个 xls 文件，内容如下

	int	char	float
0	1	NaN	1.1
1	2	NaN	2.2
2	3	NaN	3.3
3	4	d	NaN
4	5	e	NaN
5	6	NaN	6.6
6	7	NaN	7.7

import pandas as pd


demo_data = pd.read_excel('./data/demo.xls')

demo_data

	int	char	float
0	1	NaN	1.1
1	2	NaN	2.2
2	3	NaN	3.3
3	4	d	NaN
4	5	e	NaN
5	6	NaN	6.6
6	7	NaN	7.7

Pandas 缺失值判断函数 isnull()、isnull().any()、isnull().sum()

isnull() 返回一个 bool 值的 dataframe
isnull().any() 判断特征列是否存在空值
isnull().sum() 获取特征列空值数目

demo_data.isnull()

	int	char	float
0	False	True	False
1	False	True	False
2	False	True	False
3	False	False	True
4	False	False	True
5	False	True	False
6	False	True	False

demo_data.isnull().any()

int      False
char      True
float     True
dtype: bool

demo_data.isnull().sum()

int      0
char     5
float    2
dtype: int64

Pandas 缺失值填充函数 fillna()

demo_data['char'] = demo_data['char'].fillna('A')
demo_data['float'] = demo_data['float'].fillna('5.5')

demo_data

	int	char	float
0	1	A	1.1
1	2	A	2.2
2	3	A	3.3
3	4	d	5.5
4	5	e	5.5
5	6	A	6.6
6	7	A	7.7

Pandas map() 映射函数

map 里面放入字典参数，将对应的键替换成其值，注意 map 需要将所有值全部替换，否则会报错

try:
    demo_data['char'] = demo_data['char'].map({'A': 0, 'd': 1}).astype(int)
except ValueError:
    print('Error rising:Cannot convert non-finite values (NA or inf) to integer')
finally:
    demo_data['char'] = demo_data['char'].map({'A': 0, 'd': 1, 'e': 2}).astype(int)
    print('Mapping finished')

Error rising:Cannot convert non-finite values (NA or inf) to integer
Mapping finished

demo_data

	int	char	float
0	1	0	1.1
1	2	0	2.2
2	3	0	3.3
3	4	1	5.5
4	5	2	5.5
5	6	0	6.6
6	7	0	7.7

Pandas sample() 数据洗牌

sample 用于在原数据中提取数据，并进行洗牌操作，frac 代表提取的比例，为 1 表示 $100 ％$

shuffle_data_1 = demo_data.sample(frac = 1)

shuffle_data_1

	int	char	float
1	2	0	2.2
0	1	0	1.1
2	3	0	3.3
3	4	1	5.5
5	6	0	6.6
4	5	2	5.5
6	7	0	7.7

shuffle_data_2 = demo_data.sample(frac = 5 / 7)

shuffle_data_2

	int	float
6	7	7.7
0	1	1.1
2	3	3.3
5	6	6.6
1	2	2.2

Pandas drop() 删除数据

drop() 不改变原有数据，返回另一个 dataframe，使用 axis 可以指定行、列

demo_data = demo_data.drop(['char'], axis=1)

demo_data

	int	float
0	1	1.1
1	2	2.2
2	3	3.3
3	4	5.5
4	5	5.5
5	6	6.6
6	7	7.7

Pandas values 转换为 ndarray

经过上面处理，demo_data 中 data_frame 全部变成数字，使用 .values 将 data_frame 转换为 ndarray

nd_array = demo_data.values

print('nd_array:\n', nd_array,
      '\nnd_array type:', type(nd_array))

nd_array:
 [[1 1.1]
 [2 2.2]
 [3 3.3]
 [4 '5.5']
 [5 '5.5']
 [6 6.6]
 [7 7.7]] 
nd_array type:

正式开始

导入必要包

import numpy
import pandas as pd
import tensorflow as tf
import urllib.request
from sklearn import preprocessing
import matplotlib.pyplot as plt
import os
import datetime


tf.__version__

'2.0.0'

数据下载

data_url = 'http://biostat.mc.vanderbilt.edu/wiki/pub/Main/DataSets/titanic3.xls'

data_file = './data/titanic3.xls'

if not os.path.exists(data_file):
    operation = urllib.request.urlretrieve(data_url, data_file)
    print('downloading from %s' % data_url)
else:
    print('titanic3.xls is exists in the data directory')

titanic3.xls is exists in the data directory

读取数据

从数据摘要中发现 count 行的每一列数据不等，说明数据具有缺失项

dataframe = pd.read_excel(data_file)
dataframe.describe()

	pclass	survived	age	sibsp	parch	fare	body
count	1309.000000	1309.000000	1046.000000	1309.000000	1309.000000	1308.000000	121.000000
mean	2.294882	0.381971	29.881135	0.498854	0.385027	33.295479	160.809917
std	0.837836	0.486055	14.413500	1.041658	0.865560	51.758668	97.696922
min	1.000000	0.000000	0.166700	0.000000	0.000000	0.000000	1.000000
25%	2.000000	0.000000	21.000000	0.000000	0.000000	7.895800	72.000000
50%	3.000000	0.000000	28.000000	0.000000	0.000000	14.454200	155.000000
75%	3.000000	1.000000	39.000000	1.000000	0.000000	31.275000	256.000000
max	3.000000	1.000000	80.000000	8.000000	9.000000	512.329200	328.000000

dataframe

	pclass	survived	name	sex	age	sibsp	parch	ticket	fare	cabin	embarked	boat	body	home.dest
0	1	1	Allen, Miss. Elisabeth Walton	female	29.0000	0	0	24160	211.3375	B5	S	2	NaN	St Louis, MO
1	1	1	Allison, Master. Hudson Trevor	male	0.9167	1	2	113781	151.5500	C22 C26	S	11	NaN	Montreal, PQ / Chesterville, ON
2	1	0	Allison, Miss. Helen Loraine	female	2.0000	1	2	113781	151.5500	C22 C26	S	NaN	NaN	Montreal, PQ / Chesterville, ON
3	1	0	Allison, Mr. Hudson Joshua Creighton	male	30.0000	1	2	113781	151.5500	C22 C26	S	NaN	135.0	Montreal, PQ / Chesterville, ON
4	1	0	Allison, Mrs. Hudson J C (Bessie Waldo Daniels)	female	25.0000	1	2	113781	151.5500	C22 C26	S	NaN	NaN	Montreal, PQ / Chesterville, ON
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
1304	3	0	Zabour, Miss. Hileni	female	14.5000	1	0	2665	14.4542	NaN	C	NaN	328.0	NaN
1305	3	0	Zabour, Miss. Thamine	female	NaN	1	0	2665	14.4542	NaN	C	NaN	NaN	NaN
1306	3	0	Zakarian, Mr. Mapriededer	male	26.5000	0	0	2656	7.2250	NaN	C	NaN	304.0	NaN
1307	3	0	Zakarian, Mr. Ortin	male	27.0000	0	0	2670	7.2250	NaN	C	NaN	NaN	NaN
1308	3	0	Zimmerman, Mr. Leo	male	29.0000	0	0	315082	7.8750	NaN	S	NaN	NaN	NaN

1309 rows × 14 columns

数据预处理

去掉了 ticked、cabin，将 age、fare，空值用其列均值代替，sex 用0，1代替，embarked 用 S 代替，在将其所在字符值转为数字
注意：你应使用 .copy() 函数来防止严重警告⚠
selected_dataframe = dataframe[selected_cols].copy() ok
selected_dataframe = dataframe[selected_cols] not recommand

selected_cols = ['survived', 'name', 'pclass', 'sex', 'age', 'sibsp', 'parch', 'fare', 'embarked']
selected_dataframe = dataframe[selected_cols].copy()

空值填充

age_mean_value = selected_dataframe['age'].mean()
selected_dataframe['age'] = selected_dataframe['age'].fillna(age_mean_value)

fare_mean_value = selected_dataframe['fare'].mean()
selected_dataframe['fare'] = selected_dataframe['fare'].fillna(fare_mean_value)

selected_dataframe['embarked'] = selected_dataframe['embarked'].fillna('S')

selected_dataframe.describe()

	survived	pclass	age	sibsp	parch	fare
count	1309.000000	1309.000000	1309.000000	1309.000000	1309.000000	1309.000000
mean	0.381971	2.294882	29.881135	0.498854	0.385027	33.295479
std	0.486055	0.837836	12.883199	1.041658	0.865560	51.738879
min	0.000000	1.000000	0.166700	0.000000	0.000000	0.000000
25%	0.000000	2.000000	22.000000	0.000000	0.000000	7.895800
50%	0.000000	3.000000	29.881135	0.000000	0.000000	14.454200
75%	1.000000	3.000000	35.000000	1.000000	0.000000	31.275000
max	1.000000	3.000000	80.000000	8.000000	9.000000	512.329200

数字映射

selected_dataframe['sex'] = selected_dataframe['sex'].map({'female': 0, 'male': 1}).astype(int)
selected_dataframe['embarked'] = selected_dataframe['embarked'].map({'C': 0, 'Q': 1, 'S': 2}).astype(int)

创建训练数据集与标签

删除 name 列

selected_dataframe = selected_dataframe.drop(['name'], axis=1)

selected_dataframe[:3]

	survived	pclass	sex	age	sibsp	parch	fare	embarked
0	1	1	0	29.0000	0	0	211.3375	2
1	1	1	1	0.9167	1	2	151.5500	2
2	0	1	0	2.0000	1	2	151.5500	2

features 代表特征，第 1 到最后一列
label 代表标签，第 0 列

ndarray_data = selected_dataframe.values
features = ndarray_data[:, 1:]
label = ndarray_data[:, 0]

print('features:\n', features,
      '\nlabel:', label)

features:
 [[  1.       0.      29.     ...   0.     211.3375   2.    ]
 [  1.       1.       0.9167 ...   2.     151.55     2.    ]
 [  1.       0.       2.     ...   2.     151.55     2.    ]
 ...
 [  3.       1.      26.5    ...   0.       7.225    0.    ]
 [  3.       1.      27.     ...   0.       7.225    0.    ]
 [  3.       1.      29.     ...   0.       7.875    2.    ]] 
label: [1. 1. 0. ... 0. 0. 0.]

数据标准化

minmax_scale = preprocessing.MinMaxScaler(feature_range=(0, 1))
norm_features = minmax_scale.fit_transform(features)

print('norm_features:\n', norm_features,
      '\nlabel:', label)

norm_features:
 [[0.         0.         0.36116884 ... 0.         0.41250333 1.        ]
 [0.         1.         0.00939458 ... 0.22222222 0.2958059  1.        ]
 [0.         0.         0.0229641  ... 0.22222222 0.2958059  1.        ]
 ...
 [1.         1.         0.32985358 ... 0.         0.01410226 0.        ]
 [1.         1.         0.33611663 ... 0.         0.01410226 0.        ]
 [1.         1.         0.36116884 ... 0.         0.01537098 1.        ]] 
label: [1. 1. 0. ... 0. 0. 0.]

代码重构定义数据预处理函数

def prepare_data(df_data):
    df = df_data.drop(['name'], axis=1)
    age_mean = df['age'].mean()
    df['age'] = df['age'].fillna(age_mean)
    fare_mean = df['fare'].mean()
    df['fare'] = df['fare'].fillna(fare_mean)
    df['sex'] = df['sex'].map({'female':0, 'male':1}).astype(int)
    df['embarked'] = df['embarked'].fillna('S')
    df['embarked'] = df['embarked'].map({'C':0, 'Q':1, 'S':2}).astype(int)

    ndarray_data = df.values

    features = ndarray_data[:, 1:]
    label = ndarray_data[:, 0]

    minmax_scale = preprocessing.MinMaxScaler(feature_range=(0, 1))
    norm_features = minmax_scale.fit_transform(features)

    return norm_features, label

数据读取并洗牌

dataframe = pd.read_excel('./data/titanic3.xls')
selected_cols= ['survived', 'name', 'pclass', 'sex', 'age', 'sibsp', 'parch', 'fare', 'embarked']
selected_dataframe = dataframe[selected_cols].copy()
selected_dataframe = selected_dataframe.sample(frac=1)

x_data, y_data = prepare_data(selected_dataframe)

train_size = int(len(x_data) * 0.8)

x_train = x_data[:train_size]
y_train = y_data[:train_size]

x_test = x_data[train_size:]
y_test = y_data[train_size:]

model = tf.keras.models.Sequential([
    tf.keras.layers.Dense(units=64,
                          input_dim=7,
                          use_bias=True,
                          kernel_initializer='uniform',
                          bias_initializer='zeros',
                          activation='relu'),
    tf.keras.layers.Dropout(rate=0.3),
    tf.keras.layers.Dense(units=32, activation='sigmoid'),
    tf.keras.layers.Dropout(rate=0.3),
    tf.keras.layers.Dense(units=1, activation='sigmoid')
])

model.summary()

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
dense (Dense)                (None, 64)                512       
_________________________________________________________________
dropout (Dropout)            (None, 64)                0         
_________________________________________________________________
dense_1 (Dense)              (None, 32)                2080      
_________________________________________________________________
dropout_1 (Dropout)          (None, 32)                0         
_________________________________________________________________
dense_2 (Dense)              (None, 1)                 33        
=================================================================
Total params: 2,625
Trainable params: 2,625
Non-trainable params: 0
_________________________________________________________________

model.compile(optimizer=tf.keras.optimizers.Adam(0.003),
              loss='binary_crossentropy',
              metrics=['accuracy'])

train_history = model.fit(
                    x=x_train,
                    y=y_train,
                    validation_split=0.2,
                    epochs=100,
                    batch_size=40,
                    verbose=1)

Train on 837 samples, validate on 210 samples
Epoch 1/100
837/837 [==============================] - 2s 2ms/sample - loss: 0.6968 - accuracy: 0.5412 - val_loss: 0.5988 - val_accuracy: 0.6571
Epoch 2/100
837/837 [==============================] - 0s 147us/sample - loss: 0.6342 - accuracy: 0.6404 - val_loss: 0.5504 - val_accuracy: 0.7571
Epoch 3/100
837/837 [==============================] - 0s 153us/sample - loss: 0.5555 - accuracy: 0.7276 - val_loss: 0.4904 - val_accuracy: 0.8143
Epoch 4/100
837/837 [==============================] - 0s 148us/sample - loss: 0.5088 - accuracy: 0.7766 - val_loss: 0.4638 - val_accuracy: 0.8095
Epoch 5/100
837/837 [==============================] - 0s 137us/sample - loss: 0.4948 - accuracy: 0.7814 - val_loss: 0.4526 - val_accuracy: 0.8095
Epoch 6/100
837/837 [==============================] - 0s 124us/sample - loss: 0.4832 - accuracy: 0.7897 - val_loss: 0.4526 - val_accuracy: 0.8000
Epoch 7/100
837/837 [==============================] - 0s 140us/sample - loss: 0.4672 - accuracy: 0.7861 - val_loss: 0.4508 - val_accuracy: 0.8000
Epoch 8/100
837/837 [==============================] - 0s 142us/sample - loss: 0.4707 - accuracy: 0.7897 - val_loss: 0.4431 - val_accuracy: 0.8190
Epoch 9/100
837/837 [==============================] - 0s 117us/sample - loss: 0.4760 - accuracy: 0.8005 - val_loss: 0.4452 - val_accuracy: 0.8000
Epoch 10/100
837/837 [==============================] - 0s 121us/sample - loss: 0.4568 - accuracy: 0.8017 - val_loss: 0.4414 - val_accuracy: 0.7952
Epoch 11/100
837/837 [==============================] - 0s 132us/sample - loss: 0.4533 - accuracy: 0.8100 - val_loss: 0.4473 - val_accuracy: 0.7952
Epoch 12/100
837/837 [==============================] - 0s 122us/sample - loss: 0.4624 - accuracy: 0.7933 - val_loss: 0.4527 - val_accuracy: 0.8000
Epoch 13/100
837/837 [==============================] - 0s 119us/sample - loss: 0.4452 - accuracy: 0.8088 - val_loss: 0.4455 - val_accuracy: 0.8000
Epoch 14/100
837/837 [==============================] - 0s 122us/sample - loss: 0.4564 - accuracy: 0.7993 - val_loss: 0.4430 - val_accuracy: 0.7952
Epoch 15/100
837/837 [==============================] - 0s 161us/sample - loss: 0.4722 - accuracy: 0.8005 - val_loss: 0.4404 - val_accuracy: 0.8000
Epoch 16/100
837/837 [==============================] - 0s 141us/sample - loss: 0.4660 - accuracy: 0.8065 - val_loss: 0.4444 - val_accuracy: 0.8048
Epoch 17/100
837/837 [==============================] - 0s 124us/sample - loss: 0.4551 - accuracy: 0.8196 - val_loss: 0.4392 - val_accuracy: 0.8000
Epoch 18/100
837/837 [==============================] - 0s 118us/sample - loss: 0.4589 - accuracy: 0.8053 - val_loss: 0.4472 - val_accuracy: 0.8000
Epoch 19/100
837/837 [==============================] - 0s 116us/sample - loss: 0.4559 - accuracy: 0.8029 - val_loss: 0.4402 - val_accuracy: 0.8048
Epoch 20/100
837/837 [==============================] - 0s 122us/sample - loss: 0.4479 - accuracy: 0.8124 - val_loss: 0.4398 - val_accuracy: 0.7952
Epoch 21/100
837/837 [==============================] - 0s 122us/sample - loss: 0.4510 - accuracy: 0.8148 - val_loss: 0.4373 - val_accuracy: 0.8000
Epoch 22/100
837/837 [==============================] - 0s 126us/sample - loss: 0.4537 - accuracy: 0.8065 - val_loss: 0.4361 - val_accuracy: 0.8048
Epoch 23/100
837/837 [==============================] - 0s 119us/sample - loss: 0.4591 - accuracy: 0.8088 - val_loss: 0.4423 - val_accuracy: 0.7905
Epoch 24/100
837/837 [==============================] - 0s 119us/sample - loss: 0.4482 - accuracy: 0.8088 - val_loss: 0.4429 - val_accuracy: 0.7952
Epoch 25/100
837/837 [==============================] - 0s 121us/sample - loss: 0.4556 - accuracy: 0.8053 - val_loss: 0.4376 - val_accuracy: 0.8000
Epoch 26/100
837/837 [==============================] - 0s 128us/sample - loss: 0.4536 - accuracy: 0.8112 - val_loss: 0.4382 - val_accuracy: 0.8000
Epoch 27/100
837/837 [==============================] - 0s 119us/sample - loss: 0.4537 - accuracy: 0.7969 - val_loss: 0.4483 - val_accuracy: 0.8000
Epoch 28/100
837/837 [==============================] - 0s 120us/sample - loss: 0.4420 - accuracy: 0.8148 - val_loss: 0.4442 - val_accuracy: 0.7857
Epoch 29/100
837/837 [==============================] - 0s 125us/sample - loss: 0.4462 - accuracy: 0.8053 - val_loss: 0.4371 - val_accuracy: 0.7905
Epoch 30/100
837/837 [==============================] - 0s 125us/sample - loss: 0.4550 - accuracy: 0.8124 - val_loss: 0.4387 - val_accuracy: 0.7905
Epoch 31/100
837/837 [==============================] - 0s 139us/sample - loss: 0.4421 - accuracy: 0.8088 - val_loss: 0.4406 - val_accuracy: 0.7857
Epoch 32/100
837/837 [==============================] - 0s 122us/sample - loss: 0.4525 - accuracy: 0.8112 - val_loss: 0.4384 - val_accuracy: 0.7905
Epoch 33/100
837/837 [==============================] - 0s 126us/sample - loss: 0.4459 - accuracy: 0.8100 - val_loss: 0.4384 - val_accuracy: 0.7905
Epoch 34/100
837/837 [==============================] - 0s 133us/sample - loss: 0.4338 - accuracy: 0.8065 - val_loss: 0.4442 - val_accuracy: 0.7857
Epoch 35/100
837/837 [==============================] - 0s 143us/sample - loss: 0.4419 - accuracy: 0.8065 - val_loss: 0.4405 - val_accuracy: 0.7905
Epoch 36/100
837/837 [==============================] - 0s 137us/sample - loss: 0.4461 - accuracy: 0.8053 - val_loss: 0.4362 - val_accuracy: 0.7857
Epoch 37/100
837/837 [==============================] - 0s 118us/sample - loss: 0.4414 - accuracy: 0.8148 - val_loss: 0.4479 - val_accuracy: 0.7810
Epoch 38/100
837/837 [==============================] - 0s 120us/sample - loss: 0.4382 - accuracy: 0.8136 - val_loss: 0.4365 - val_accuracy: 0.7905
Epoch 39/100
837/837 [==============================] - 0s 125us/sample - loss: 0.4356 - accuracy: 0.8184 - val_loss: 0.4488 - val_accuracy: 0.7810
Epoch 40/100
837/837 [==============================] - 0s 143us/sample - loss: 0.4383 - accuracy: 0.8184 - val_loss: 0.4375 - val_accuracy: 0.7905
Epoch 41/100
837/837 [==============================] - 0s 150us/sample - loss: 0.4347 - accuracy: 0.8124 - val_loss: 0.4451 - val_accuracy: 0.7810
Epoch 42/100
837/837 [==============================] - 0s 143us/sample - loss: 0.4573 - accuracy: 0.8112 - val_loss: 0.4411 - val_accuracy: 0.7857
Epoch 43/100
837/837 [==============================] - 0s 119us/sample - loss: 0.4358 - accuracy: 0.8136 - val_loss: 0.4379 - val_accuracy: 0.7905
Epoch 44/100
837/837 [==============================] - 0s 164us/sample - loss: 0.4453 - accuracy: 0.8160 - val_loss: 0.4458 - val_accuracy: 0.7810
Epoch 45/100
837/837 [==============================] - 0s 126us/sample - loss: 0.4401 - accuracy: 0.8076 - val_loss: 0.4405 - val_accuracy: 0.7952
Epoch 46/100
837/837 [==============================] - 0s 142us/sample - loss: 0.4364 - accuracy: 0.8160 - val_loss: 0.4465 - val_accuracy: 0.7810
Epoch 47/100
837/837 [==============================] - 0s 129us/sample - loss: 0.4311 - accuracy: 0.8184 - val_loss: 0.4386 - val_accuracy: 0.8000
Epoch 48/100
837/837 [==============================] - 0s 117us/sample - loss: 0.4377 - accuracy: 0.8160 - val_loss: 0.4444 - val_accuracy: 0.7857
Epoch 49/100
837/837 [==============================] - 0s 133us/sample - loss: 0.4546 - accuracy: 0.7957 - val_loss: 0.4432 - val_accuracy: 0.7810
Epoch 50/100
837/837 [==============================] - 0s 145us/sample - loss: 0.4403 - accuracy: 0.8208 - val_loss: 0.4436 - val_accuracy: 0.7905
Epoch 51/100
837/837 [==============================] - 0s 144us/sample - loss: 0.4259 - accuracy: 0.8148 - val_loss: 0.4374 - val_accuracy: 0.7952
Epoch 52/100
837/837 [==============================] - 0s 155us/sample - loss: 0.4300 - accuracy: 0.8160 - val_loss: 0.4411 - val_accuracy: 0.7857
Epoch 53/100
837/837 [==============================] - 0s 160us/sample - loss: 0.4381 - accuracy: 0.8136 - val_loss: 0.4432 - val_accuracy: 0.7905
Epoch 54/100
837/837 [==============================] - 0s 136us/sample - loss: 0.4290 - accuracy: 0.8256 - val_loss: 0.4414 - val_accuracy: 0.7810
Epoch 55/100
837/837 [==============================] - 0s 148us/sample - loss: 0.4360 - accuracy: 0.8160 - val_loss: 0.4385 - val_accuracy: 0.7952
Epoch 56/100
837/837 [==============================] - 0s 114us/sample - loss: 0.4364 - accuracy: 0.8232 - val_loss: 0.4415 - val_accuracy: 0.7810
Epoch 57/100
837/837 [==============================] - 0s 127us/sample - loss: 0.4364 - accuracy: 0.8076 - val_loss: 0.4397 - val_accuracy: 0.7952
Epoch 58/100
837/837 [==============================] - 0s 122us/sample - loss: 0.4370 - accuracy: 0.8148 - val_loss: 0.4378 - val_accuracy: 0.7905
Epoch 59/100
837/837 [==============================] - 0s 139us/sample - loss: 0.4435 - accuracy: 0.8088 - val_loss: 0.4444 - val_accuracy: 0.7810
Epoch 60/100
837/837 [==============================] - 0s 122us/sample - loss: 0.4354 - accuracy: 0.8172 - val_loss: 0.4372 - val_accuracy: 0.7952
Epoch 61/100
837/837 [==============================] - 0s 137us/sample - loss: 0.4375 - accuracy: 0.8112 - val_loss: 0.4420 - val_accuracy: 0.7857
Epoch 62/100
837/837 [==============================] - 0s 135us/sample - loss: 0.4307 - accuracy: 0.8136 - val_loss: 0.4392 - val_accuracy: 0.7905
Epoch 63/100
837/837 [==============================] - 0s 158us/sample - loss: 0.4362 - accuracy: 0.8160 - val_loss: 0.4406 - val_accuracy: 0.7952
Epoch 64/100
837/837 [==============================] - 0s 124us/sample - loss: 0.4405 - accuracy: 0.8124 - val_loss: 0.4440 - val_accuracy: 0.7810
Epoch 65/100
837/837 [==============================] - 0s 119us/sample - loss: 0.4298 - accuracy: 0.8232 - val_loss: 0.4386 - val_accuracy: 0.8000
Epoch 66/100
837/837 [==============================] - 0s 108us/sample - loss: 0.4284 - accuracy: 0.8065 - val_loss: 0.4446 - val_accuracy: 0.7905
Epoch 67/100
837/837 [==============================] - 0s 129us/sample - loss: 0.4340 - accuracy: 0.8148 - val_loss: 0.4438 - val_accuracy: 0.7905
Epoch 68/100
837/837 [==============================] - 0s 114us/sample - loss: 0.4389 - accuracy: 0.8088 - val_loss: 0.4381 - val_accuracy: 0.8000
Epoch 69/100
837/837 [==============================] - 0s 124us/sample - loss: 0.4338 - accuracy: 0.8220 - val_loss: 0.4386 - val_accuracy: 0.8000
Epoch 70/100
837/837 [==============================] - 0s 115us/sample - loss: 0.4323 - accuracy: 0.8184 - val_loss: 0.4423 - val_accuracy: 0.7952
Epoch 71/100
837/837 [==============================] - 0s 122us/sample - loss: 0.4237 - accuracy: 0.8232 - val_loss: 0.4406 - val_accuracy: 0.7905
Epoch 72/100
837/837 [==============================] - 0s 137us/sample - loss: 0.4355 - accuracy: 0.8208 - val_loss: 0.4437 - val_accuracy: 0.7905
Epoch 73/100
837/837 [==============================] - 0s 116us/sample - loss: 0.4362 - accuracy: 0.8196 - val_loss: 0.4364 - val_accuracy: 0.7905
Epoch 74/100
837/837 [==============================] - 0s 128us/sample - loss: 0.4293 - accuracy: 0.8196 - val_loss: 0.4445 - val_accuracy: 0.7810
Epoch 75/100
837/837 [==============================] - 0s 124us/sample - loss: 0.4252 - accuracy: 0.8184 - val_loss: 0.4400 - val_accuracy: 0.7905
Epoch 76/100
837/837 [==============================] - 0s 119us/sample - loss: 0.4335 - accuracy: 0.8256 - val_loss: 0.4470 - val_accuracy: 0.7810
Epoch 77/100
837/837 [==============================] - 0s 119us/sample - loss: 0.4284 - accuracy: 0.8184 - val_loss: 0.4384 - val_accuracy: 0.8000
Epoch 78/100
837/837 [==============================] - 0s 147us/sample - loss: 0.4398 - accuracy: 0.8136 - val_loss: 0.4412 - val_accuracy: 0.7905
Epoch 79/100
837/837 [==============================] - 0s 129us/sample - loss: 0.4339 - accuracy: 0.8160 - val_loss: 0.4454 - val_accuracy: 0.7810
Epoch 80/100
837/837 [==============================] - 0s 127us/sample - loss: 0.4286 - accuracy: 0.8160 - val_loss: 0.4397 - val_accuracy: 0.7905
Epoch 81/100
837/837 [==============================] - 0s 120us/sample - loss: 0.4315 - accuracy: 0.8220 - val_loss: 0.4393 - val_accuracy: 0.7905
Epoch 82/100
837/837 [==============================] - 0s 138us/sample - loss: 0.4263 - accuracy: 0.8184 - val_loss: 0.4415 - val_accuracy: 0.7905
Epoch 83/100
837/837 [==============================] - 0s 136us/sample - loss: 0.4298 - accuracy: 0.8208 - val_loss: 0.4405 - val_accuracy: 0.8048
Epoch 84/100
837/837 [==============================] - 0s 129us/sample - loss: 0.4341 - accuracy: 0.8112 - val_loss: 0.4377 - val_accuracy: 0.7952
Epoch 85/100
837/837 [==============================] - 0s 127us/sample - loss: 0.4325 - accuracy: 0.8100 - val_loss: 0.4432 - val_accuracy: 0.8000
Epoch 86/100
837/837 [==============================] - 0s 132us/sample - loss: 0.4277 - accuracy: 0.8124 - val_loss: 0.4415 - val_accuracy: 0.7857
Epoch 87/100
837/837 [==============================] - 0s 113us/sample - loss: 0.4274 - accuracy: 0.8196 - val_loss: 0.4427 - val_accuracy: 0.7905
Epoch 88/100
837/837 [==============================] - 0s 112us/sample - loss: 0.4243 - accuracy: 0.8280 - val_loss: 0.4400 - val_accuracy: 0.7905
Epoch 89/100
837/837 [==============================] - 0s 118us/sample - loss: 0.4280 - accuracy: 0.8220 - val_loss: 0.4418 - val_accuracy: 0.7952
Epoch 90/100
837/837 [==============================] - 0s 112us/sample - loss: 0.4340 - accuracy: 0.8208 - val_loss: 0.4409 - val_accuracy: 0.8000
Epoch 91/100
837/837 [==============================] - 0s 121us/sample - loss: 0.4298 - accuracy: 0.8136 - val_loss: 0.4403 - val_accuracy: 0.8000
Epoch 92/100
837/837 [==============================] - 0s 122us/sample - loss: 0.4275 - accuracy: 0.8208 - val_loss: 0.4409 - val_accuracy: 0.7952
Epoch 93/100
837/837 [==============================] - 0s 136us/sample - loss: 0.4228 - accuracy: 0.8244 - val_loss: 0.4394 - val_accuracy: 0.8095
Epoch 94/100
837/837 [==============================] - 0s 119us/sample - loss: 0.4313 - accuracy: 0.8208 - val_loss: 0.4434 - val_accuracy: 0.8000
Epoch 95/100
837/837 [==============================] - 0s 130us/sample - loss: 0.4277 - accuracy: 0.8196 - val_loss: 0.4365 - val_accuracy: 0.8095
Epoch 96/100
837/837 [==============================] - 0s 118us/sample - loss: 0.4273 - accuracy: 0.8220 - val_loss: 0.4383 - val_accuracy: 0.8000
Epoch 97/100
837/837 [==============================] - 0s 113us/sample - loss: 0.4311 - accuracy: 0.8124 - val_loss: 0.4373 - val_accuracy: 0.8095
Epoch 98/100
837/837 [==============================] - 0s 123us/sample - loss: 0.4221 - accuracy: 0.8327 - val_loss: 0.4419 - val_accuracy: 0.7952
Epoch 99/100
837/837 [==============================] - 0s 124us/sample - loss: 0.4378 - accuracy: 0.8196 - val_loss: 0.4380 - val_accuracy: 0.8048
Epoch 100/100
837/837 [==============================] - 0s 122us/sample - loss: 0.4238 - accuracy: 0.8232 - val_loss: 0.4451 - val_accuracy: 0.7857

训练可视化

fig = plt.gcf()
fig.set_size_inches(10, 5)
ax1 = fig.add_subplot(111)
ax1.set_title('Train and Validation Picture')
ax1.set_ylabel('Loss value')
line1, = ax1.plot(train_history.history['loss'], color=(0.5, 0.5, 1.0), label='Loss train')
line2, = ax1.plot(train_history.history['val_loss'], color=(0.5, 1.0, 0.5), label='Loss valid')
ax2 = ax1.twinx()
ax2.set_ylabel('Accuracy value')
line3, = ax2.plot(train_history.history['accuracy'], color=(0.5, 0.5, 0.5), label='Accuracy train')
line4, = ax2.plot(train_history.history['val_accuracy'], color=(1, 0, 0), label='Accuracy valid')
plt.legend(handles=(line1, line2, line3, line4), loc='best')
plt.show()

测试模型

test_loss, test_acc = model.evaluate(x_test, y_test, verbose=2)
print('test_loss:', test_loss,
      '\ntest_acc:', test_acc,
      '\nmetrics_names:', model.metrics_names)

262/1 - 0s - loss: 0.3581 - accuracy: 0.7672
test_loss: 0.48995060536242624 
test_acc: 0.76717556 
metrics_names: ['loss', 'accuracy']

使用 Jack & Rose 测试

Jack_info = [0, 'Jack', 3, 'male', 23, 1, 0, 5.0000, 'S']
Rose_info = [1, 'Rose', 1, 'female', 20, 1, 0, 100.0000, 'S']

new_passenger_pd = pd.DataFrame([Jack_info, Rose_info], columns=selected_cols)
all_passenger_pd = selected_dataframe.append(new_passenger_pd)

pred = model.predict(prepare_data(all_passenger_pd)[0])

print('Rose survived probability:', pred[-1:][0][0],
      '\nJack survived probability:', pred[-2:][0][0])

Rose survived probability: 0.96711206 
Jack survived probability: 0.12514974

数据插入最后一列生存概率

all_passenger_pd.insert(len(all_passenger_pd.columns), 'surv_prob', pred)

all_passenger_pd

	survived	name	pclass	sex	age	sibsp	parch	fare	embarked	surv_prob
75	0	Colley, Mr. Edward Pomeroy	1	male	47.0	0	0	25.5875	S	0.221973
321	0	Wright, Mr. George	1	male	62.0	0	0	26.5500	S	0.194789
712	0	Celotti, Mr. Francesco	3	male	24.0	0	0	8.0500	S	0.130646
345	0	Berriman, Mr. William John	2	male	23.0	0	0	13.0000	S	0.211936
1298	0	Wittevrongel, Mr. Camille	3	male	36.0	0	0	9.5000	S	0.106074
...	...	...	...	...	...	...	...	...	...	...
10	0	Astor, Col. John Jacob	1	male	47.0	1	0	227.5250	C	0.213140
434	1	Hart, Miss. Eva Miriam	2	female	7.0	0	2	26.2500	S	0.881991
690	0	Brobeck, Mr. Karl Rudolf	3	male	22.0	0	0	7.7958	S	0.136104
0	0	Jack	3	male	23.0	1	0	5.0000	S	0.125150
1	1	Rose	1	female	20.0	1	0	100.0000	S	0.967112

1311 rows × 10 columns

form =pd.DataFrame(columns=[column for column in all_passenger_pd], data=all_passenger_pd)
form.to_excel('./data/result.xls', encoding='utf-8', index=None, header=True)

加入回调

def prepare_data(df_data):
    df = df_data.drop(['name'], axis=1)
    age_mean = df['age'].mean()
    df['age'] = df['age'].fillna(age_mean)
    fare_mean = df['fare'].mean()
    df['fare'] = df['fare'].fillna(fare_mean)
    df['sex'] = df['sex'].map({'female':0, 'male':1}).astype(int)
    df['embarked'] = df['embarked'].fillna('S')
    df['embarked'] = df['embarked'].map({'C':0, 'Q':1, 'S':2}).astype(int)

    ndarray_data = df.values

    features = ndarray_data[:, 1:]
    label = ndarray_data[:, 0]

    minmax_scale = preprocessing.MinMaxScaler(feature_range=(0, 1))
    norm_features = minmax_scale.fit_transform(features)

    return norm_features, label

dataframe = pd.read_excel('./data/titanic3.xls')
selected_cols= ['survived', 'name', 'pclass', 'sex', 'age', 'sibsp', 'parch', 'fare', 'embarked']
selected_dataframe = dataframe[selected_cols].copy()
selected_dataframe = selected_dataframe.sample(frac=1)

x_data, y_data = prepare_data(selected_dataframe)

train_size = int(len(x_data) * 0.8)

x_train = x_data[:train_size]
y_train = y_data[:train_size]

x_test = x_data[train_size:]
y_test = y_data[train_size:]

model = tf.keras.models.Sequential([
    tf.keras.layers.Dense(units=256,
                          input_dim=7,
                          use_bias=True,
                          kernel_initializer='uniform',
                          bias_initializer='zeros',
                          activation='relu'),
    tf.keras.layers.Dropout(rate=0.3),
    tf.keras.layers.Dense(units=128, activation='sigmoid'),
    tf.keras.layers.Dropout(rate=0.3),
    tf.keras.layers.Dense(units=64, activation='sigmoid'),
    tf.keras.layers.Dropout(rate=0.3),
    tf.keras.layers.Dense(units=32, activation='sigmoid'),
    tf.keras.layers.Dropout(rate=0.3),
    tf.keras.layers.Dense(units=1, activation='sigmoid')
])

model.summary()

Model: "sequential_1"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
dense_3 (Dense)              (None, 256)               2048      
_________________________________________________________________
dropout_2 (Dropout)          (None, 256)               0         
_________________________________________________________________
dense_4 (Dense)              (None, 128)               32896     
_________________________________________________________________
dropout_3 (Dropout)          (None, 128)               0         
_________________________________________________________________
dense_5 (Dense)              (None, 64)                8256      
_________________________________________________________________
dropout_4 (Dropout)          (None, 64)                0         
_________________________________________________________________
dense_6 (Dense)              (None, 32)                2080      
_________________________________________________________________
dropout_5 (Dropout)          (None, 32)                0         
_________________________________________________________________
dense_7 (Dense)              (None, 1)                 33        
=================================================================
Total params: 45,313
Trainable params: 45,313
Non-trainable params: 0
_________________________________________________________________

model.compile(optimizer=tf.keras.optimizers.Adam(0.003),
              loss='binary_crossentropy',
              metrics=['accuracy'])

log_dir = os.path.join(
    'logs2.x',
    'train',
    'plugins',
    'profile',
    datetime.datetime.now().strftime('%Y-%m-%d_%H-%M-%S'))

checkpoint_path = './checkpoint2.x/Titanic.{epoch:02d}.ckpt'
if not os.path.exists('./checkpoint2.x'):
    os.mkdir('./checkpoint2.x')

callbacks = [tf.keras.callbacks.TensorBoard(log_dir=log_dir,
                         histogram_freq=2),
       tf.keras.callbacks.ModelCheckpoint(filepath=checkpoint_path,
                         save_weights_only=True,
                         verbose=1,
                         period=5)]

WARNING:tensorflow:`period` argument is deprecated. Please use `save_freq` to specify the frequency in number of samples seen.

train_history = model.fit(x=x_train, y=y_train,
                          validation_split=0.2,
                          epochs=100,
                          batch_size=40,
                          callbacks=callbacks,
                          verbose=1)

部分训练如下
Train on 837 samples, validate on 210 samples
Epoch 80/100
760/837 [==========================>...] - ETA: 0s - loss: 0.4277 - accuracy: 0.8132
Epoch 00080: saving model to ./checkpoint2.x/Titanic.80.h5
837/837 [==============================] - 0s 302us/sample - loss: 0.4353 - accuracy: 0.8112 - val_loss: 0.4639 - val_accuracy: 0.7810
Epoch 81/100
837/837 [==============================] - 0s 270us/sample - loss: 0.4455 - accuracy: 0.8017 - val_loss: 0.4768 - val_accuracy: 0.7810
Epoch 82/100
837/837 [==============================] - 0s 211us/sample - loss: 0.4376 - accuracy: 0.7993 - val_loss: 0.4654 - val_accuracy: 0.7905
Epoch 83/100
837/837 [==============================] - 0s 278us/sample - loss: 0.4377 - accuracy: 0.8065 - val_loss: 0.4703 - val_accuracy: 0.7810
Epoch 84/100
837/837 [==============================] - 0s 232us/sample - loss: 0.4368 - accuracy: 0.8160 - val_loss: 0.4631 - val_accuracy: 0.7952
Epoch 85/100
360/837 [===========>..................] - ETA: 0s - loss: 0.4669 - accuracy: 0.8056
Epoch 00085: saving model to ./checkpoint2.x/Titanic.85.h5
837/837 [==============================] - 0s 292us/sample - loss: 0.4437 - accuracy: 0.8124 - val_loss: 0.4627 - val_accuracy: 0.7810
Epoch 86/100
837/837 [==============================] - 0s 197us/sample - loss: 0.4365 - accuracy: 0.8017 - val_loss: 0.4686 - val_accuracy: 0.7905
Epoch 87/100
837/837 [==============================] - 0s 288us/sample - loss: 0.4500 - accuracy: 0.8148 - val_loss: 0.4689 - val_accuracy: 0.7857
Epoch 88/100
837/837 [==============================] - 0s 208us/sample - loss: 0.4356 - accuracy: 0.8029 - val_loss: 0.4794 - val_accuracy: 0.7905
Epoch 89/100
837/837 [==============================] - 0s 239us/sample - loss: 0.4283 - accuracy: 0.8148 - val_loss: 0.4621 - val_accuracy: 0.7857
Epoch 90/100
440/837 [==============>...............] - ETA: 0s - loss: 0.4083 - accuracy: 0.8295
Epoch 00090: saving model to ./checkpoint2.x/Titanic.90.h5
837/837 [==============================] - 0s 258us/sample - loss: 0.4359 - accuracy: 0.8172 - val_loss: 0.4736 - val_accuracy: 0.7905
Epoch 91/100
837/837 [==============================] - 0s 299us/sample - loss: 0.4365 - accuracy: 0.8053 - val_loss: 0.4658 - val_accuracy: 0.7905
Epoch 92/100
837/837 [==============================] - 0s 319us/sample - loss: 0.4376 - accuracy: 0.8148 - val_loss: 0.4696 - val_accuracy: 0.7905
Epoch 93/100
837/837 [==============================] - 0s 355us/sample - loss: 0.4375 - accuracy: 0.8005 - val_loss: 0.4698 - val_accuracy: 0.7952
Epoch 94/100
837/837 [==============================] - 0s 205us/sample - loss: 0.4384 - accuracy: 0.8005 - val_loss: 0.4682 - val_accuracy: 0.7905
Epoch 95/100
440/837 [==============>...............] - ETA: 0s - loss: 0.4514 - accuracy: 0.7909
Epoch 00095: saving model to ./checkpoint2.x/Titanic.95.h5
837/837 [==============================] - 0s 344us/sample - loss: 0.4392 - accuracy: 0.8005 - val_loss: 0.4620 - val_accuracy: 0.7952
Epoch 96/100
837/837 [==============================] - 0s 219us/sample - loss: 0.4347 - accuracy: 0.8053 - val_loss: 0.4643 - val_accuracy: 0.7857
Epoch 97/100
837/837 [==============================] - 0s 309us/sample - loss: 0.4410 - accuracy: 0.8005 - val_loss: 0.4772 - val_accuracy: 0.7905
Epoch 98/100
837/837 [==============================] - 0s 230us/sample - loss: 0.4325 - accuracy: 0.8076 - val_loss: 0.4629 - val_accuracy: 0.7857
Epoch 99/100
837/837 [==============================] - 0s 267us/sample - loss: 0.4308 - accuracy: 0.8005 - val_loss: 0.4658 - val_accuracy: 0.7857
Epoch 100/100
360/837 [===========>..................] - ETA: 0s - loss: 0.4338 - accuracy: 0.8139
Epoch 00100: saving model to ./checkpoint2.x/Titanic.100.h5
837/837 [==============================] - 0s 265us/sample - loss: 0.4314 - accuracy: 0.8124 - val_loss: 0.4623 - val_accuracy: 0.7857

fig = plt.gcf()
fig.set_size_inches(10, 5)
ax1 = fig.add_subplot(111)
ax1.set_title('Train and Validation Picture')
ax1.set_ylabel('Loss value')
line1, = ax1.plot(train_history.history['loss'], color=(0.5, 0.5, 1.0), label='Loss train')
line2, = ax1.plot(train_history.history['val_loss'], color=(0.5, 1.0, 0.5), label='Loss valid')
ax2 = ax1.twinx()
ax2.set_ylabel('Accuracy value')
line3, = ax2.plot(train_history.history['accuracy'], color=(0.5, 0.5, 0.5), label='Accuracy train')
line4, = ax2.plot(train_history.history['val_accuracy'], color=(1, 0, 0), label='Accuracy valid')
plt.legend(handles=(line1, line2, line3, line4), loc='best')
plt.show()

Jack_info = [0, 'Jack', 3, 'male', 23, 1, 0, 5.0000, 'S']
Rose_info = [1, 'Rose', 1, 'female', 20, 1, 0, 100.0000, 'S']

new_passenger_pd = pd.DataFrame([Jack_info, Rose_info], columns=selected_cols)
all_passenger_pd = selected_dataframe.append(new_passenger_pd)

pred = model.predict(prepare_data(all_passenger_pd)[0])

print('Rose survived probability:', pred[-1:][0][0],
      '\nJack survived probability:', pred[-2:][0][0])

Rose survived probability: 0.9700622 
Jack survived probability: 0.12726058

加载模型

由于只保存了网络参数，没有保存网络结构，需要重新定义网络结构（当然，由于 jupyter 的缓存效应，你大可不必重新定义，对于独立的 py 文件则需要这么做）

model = tf.keras.models.Sequential([
    tf.keras.layers.Dense(units=256,
                          input_dim=7,
                          use_bias=True,
                          kernel_initializer='uniform',
                          bias_initializer='zeros',
                          activation='relu'),
    tf.keras.layers.Dropout(rate=0.3),
    tf.keras.layers.Dense(units=128, activation='sigmoid'),
    tf.keras.layers.Dropout(rate=0.3),
    tf.keras.layers.Dense(units=64, activation='sigmoid'),
    tf.keras.layers.Dropout(rate=0.3),
    tf.keras.layers.Dense(units=32, activation='sigmoid'),
    tf.keras.layers.Dropout(rate=0.3),
    tf.keras.layers.Dense(units=1, activation='sigmoid')
])

model.load_weights('./checkpoint2.x/Titanic.100.h5')

model.compile(optimizer=tf.keras.optimizers.Adam(0.003),
              loss='binary_crossentropy',
              metrics=['accuracy'])

loss, acc = model.evaluate(x_test, y_test, verbose=2)
print('Restore model accuracy:{:5.4f}%'.format(100 * acc))

262/1 - 0s - loss: 0.5042 - accuracy: 0.8511
Restore model accuracy:85.1145%

Titanic TensorFLow 1.x Keras API 实现

导入必要的包

import numpy
import pandas as pd
import tensorflow as tf
import urllib.request
from sklearn import preprocessing
import matplotlib.pyplot as plt
import os
import datetime


tf.__version__

'1.15.2'

定义预处理函数

def prepare_data(df_data):
    df = df_data.drop(['name'], axis=1)
    age_mean = df['age'].mean()
    df['age'] = df['age'].fillna(age_mean)
    fare_mean = df['fare'].mean()
    df['fare'] = df['fare'].fillna(fare_mean)
    df['sex'] = df['sex'].map({'female':0, 'male':1}).astype(int)
    df['embarked'] = df['embarked'].fillna('S')
    df['embarked'] = df['embarked'].map({'C':0, 'Q':1, 'S':2}).astype(int)

    ndarray_data = df.values

    features = ndarray_data[:, 1:]
    label = ndarray_data[:, 0]

    minmax_scale = preprocessing.MinMaxScaler(feature_range=(0, 1))
    norm_features = minmax_scale.fit_transform(features)

    return norm_features, label

读取数据，制作数据集

dataframe = pd.read_excel('./data/titanic3.xls')
selected_cols= ['survived', 'name', 'pclass', 'sex', 'age', 'sibsp', 'parch', 'fare', 'embarked']
selected_dataframe = dataframe[selected_cols].copy()
selected_dataframe = selected_dataframe.sample(frac=1)

x_data, y_data = prepare_data(selected_dataframe)

train_size = int(len(x_data) * 0.8)

x_train = x_data[:train_size]
y_train = y_data[:train_size]

x_test = x_data[train_size:]
y_test = y_data[train_size:]

搭建模型

model = tf.keras.models.Sequential([
    tf.keras.layers.Dense(units=64,
                          input_dim=7,
                          use_bias=True,
                          kernel_initializer='uniform',
                          bias_initializer='zeros',
                          activation='relu'),
    tf.keras.layers.Dropout(rate=0.3),
    tf.keras.layers.Dense(units=32, activation='sigmoid'),
    tf.keras.layers.Dropout(rate=0.3),
    tf.keras.layers.Dense(units=1, activation='sigmoid')
])

WARNING:tensorflow:From e:\anaconda3\envs\tensorflow1.x\lib\site-packages\tensorflow_core\python\keras\initializers.py:119: calling RandomUniform.__init__ (from tensorflow.python.ops.init_ops) with dtype is deprecated and will be removed in a future version.
Instructions for updating:
Call initializer instance with the dtype argument instead of passing it to the constructor
WARNING:tensorflow:From e:\anaconda3\envs\tensorflow1.x\lib\site-packages\tensorflow_core\python\ops\resource_variable_ops.py:1630: calling BaseResourceVariable.__init__ (from tensorflow.python.ops.resource_variable_ops) with constraint is deprecated and will be removed in a future version.
Instructions for updating:
If using Keras pass *_constraint arguments to layers.

model.summary()

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
dense (Dense)                (None, 64)                512       
_________________________________________________________________
dropout (Dropout)            (None, 64)                0         
_________________________________________________________________
dense_1 (Dense)              (None, 32)                2080      
_________________________________________________________________
dropout_1 (Dropout)          (None, 32)                0         
_________________________________________________________________
dense_2 (Dense)              (None, 1)                 33        
=================================================================
Total params: 2,625
Trainable params: 2,625
Non-trainable params: 0
_________________________________________________________________

model.compile(optimizer=tf.keras.optimizers.Adam(0.003),
              loss='binary_crossentropy',
              metrics=['accuracy'])

WARNING:tensorflow:From e:\anaconda3\envs\tensorflow1.x\lib\site-packages\tensorflow_core\python\ops\nn_impl.py:183: where (from tensorflow.python.ops.array_ops) is deprecated and will be removed in a future version.
Instructions for updating:
Use tf.where in 2.0, which has the same broadcast rule as np.where

log_dir = os.path.join(
    'logs1.x',
    'train',
    'plugins',
    'profile',
    datetime.datetime.now().strftime('%Y-%m-%d_%H-%M-%S'))

checkpoint_path = './checkpoint1.x/Titanic_{epoch:02d}-{val_loss:.2f}.ckpt'


callbacks = [tf.keras.callbacks.TensorBoard(log_dir=log_dir,
                         histogram_freq=2),
       tf.keras.callbacks.ModelCheckpoint(filepath=checkpoint_path,
                         save_weights_only=True,
                         verbose=1,
                         period=5)]

WARNING:tensorflow:`period` argument is deprecated. Please use `save_freq` to specify the frequency in number of samples seen.

开始训练

train_history = model.fit(x=x_train,
              y=y_train,
              validation_split=0.2,
              epochs=100,
              batch_size=40,
              callbacks=callbacks,
              verbose=2)

部分训练如下
Train on 837 samples, validate on 210 samples
837/837 - 0s - loss: 0.4398 - acc: 0.8124 - val_loss: 0.4671 - val_acc: 0.7857
Epoch 80/100

Epoch 00080: saving model to ./checkpoint1.x/Titanic_80-0.47.ckpt
837/837 - 0s - loss: 0.4360 - acc: 0.8076 - val_loss: 0.4673 - val_acc: 0.7857
Epoch 81/100
837/837 - 0s - loss: 0.4307 - acc: 0.8005 - val_loss: 0.4703 - val_acc: 0.7905
Epoch 82/100
837/837 - 0s - loss: 0.4401 - acc: 0.7981 - val_loss: 0.4666 - val_acc: 0.8000
Epoch 83/100
837/837 - 0s - loss: 0.4311 - acc: 0.8017 - val_loss: 0.4678 - val_acc: 0.7952
Epoch 84/100
837/837 - 0s - loss: 0.4296 - acc: 0.8172 - val_loss: 0.4673 - val_acc: 0.8000
Epoch 85/100

Epoch 00085: saving model to ./checkpoint1.x/Titanic_85-0.46.ckpt
837/837 - 0s - loss: 0.4384 - acc: 0.8029 - val_loss: 0.4634 - val_acc: 0.7857
Epoch 86/100
837/837 - 0s - loss: 0.4345 - acc: 0.8076 - val_loss: 0.4666 - val_acc: 0.7905
Epoch 87/100
837/837 - 0s - loss: 0.4307 - acc: 0.8053 - val_loss: 0.4650 - val_acc: 0.8000
Epoch 88/100
837/837 - 0s - loss: 0.4394 - acc: 0.8148 - val_loss: 0.4638 - val_acc: 0.8000
Epoch 89/100
837/837 - 0s - loss: 0.4355 - acc: 0.8053 - val_loss: 0.4648 - val_acc: 0.8000
Epoch 90/100

Epoch 00090: saving model to ./checkpoint1.x/Titanic_90-0.46.ckpt
837/837 - 0s - loss: 0.4326 - acc: 0.8100 - val_loss: 0.4623 - val_acc: 0.8000
Epoch 91/100
837/837 - 0s - loss: 0.4387 - acc: 0.8029 - val_loss: 0.4658 - val_acc: 0.7905
Epoch 92/100
837/837 - 0s - loss: 0.4285 - acc: 0.8065 - val_loss: 0.4613 - val_acc: 0.7905
Epoch 93/100
837/837 - 0s - loss: 0.4355 - acc: 0.8088 - val_loss: 0.4656 - val_acc: 0.7905
Epoch 94/100
837/837 - 0s - loss: 0.4318 - acc: 0.8136 - val_loss: 0.4629 - val_acc: 0.7952
Epoch 95/100

Epoch 00095: saving model to ./checkpoint1.x/Titanic_95-0.46.ckpt
837/837 - 0s - loss: 0.4386 - acc: 0.7981 - val_loss: 0.4639 - val_acc: 0.8000
Epoch 96/100
837/837 - 0s - loss: 0.4346 - acc: 0.8041 - val_loss: 0.4647 - val_acc: 0.7857
Epoch 97/100
837/837 - 0s - loss: 0.4256 - acc: 0.8160 - val_loss: 0.4608 - val_acc: 0.8048
Epoch 98/100
837/837 - 0s - loss: 0.4357 - acc: 0.8029 - val_loss: 0.4613 - val_acc: 0.8000
Epoch 99/100
837/837 - 0s - loss: 0.4265 - acc: 0.8041 - val_loss: 0.4614 - val_acc: 0.7952
Epoch 100/100

Epoch 00100: saving model to ./checkpoint1.x/Titanic_100-0.46.ckpt
837/837 - 0s - loss: 0.4243 - acc: 0.8148 - val_loss: 0.4611 - val_acc: 0.8000

训练可视化

fig = plt.gcf()
fig.set_size_inches(10, 5)
ax1 = fig.add_subplot(111)
ax1.set_title('Train and Validation Picture')
ax1.set_ylabel('Loss value')
line1, = ax1.plot(train_history.history['loss'], color=(0.5, 0.5, 1.0), label='Loss train')
line2, = ax1.plot(train_history.history['val_loss'], color=(0.5, 1.0, 0.5), label='Loss valid')
ax2 = ax1.twinx()
ax2.set_ylabel('Accuracy value')
line3, = ax2.plot(train_history.history['acc'], color=(0.5, 0.5, 0.5), label='Accuracy train')
line4, = ax2.plot(train_history.history['val_acc'], color=(1, 0, 0), label='Accuracy valid')
plt.legend(handles=(line1, line2, line3, line4), loc='best')
plt.show()

模型预测

Jack_info = [0, 'Jack', 3, 'male', 23, 1, 0, 5.0000, 'S']
Rose_info = [1, 'Rose', 1, 'female', 20, 1, 0, 100.0000, 'S']

new_passenger_pd = pd.DataFrame([Jack_info, Rose_info], columns=selected_cols)
all_passenger_pd = selected_dataframe.append(new_passenger_pd)

pred = model.predict(prepare_data(all_passenger_pd)[0])

print('Rose survived probability:', pred[-1:][0][0],
      '\nJack survived probability:', pred[-2:][0][0])

Rose survived probability: 0.9762004 
Jack survived probability: 0.10789904

加载模型进行预测

model = tf.keras.models.Sequential([
    tf.keras.layers.Dense(units=64,
                          input_dim=7,
                          use_bias=True,
                          kernel_initializer='uniform',
                          bias_initializer='zeros',
                          activation='relu'),
    tf.keras.layers.Dropout(rate=0.3),
    tf.keras.layers.Dense(units=32, activation='sigmoid'),
    tf.keras.layers.Dropout(rate=0.3),
    tf.keras.layers.Dense(units=1, activation='sigmoid')
])

model.compile(optimizer=tf.keras.optimizers.Adam(0.003),
              loss='binary_crossentropy',
              metrics=['accuracy'])

checkpoint_dir = os.path.dirname(checkpoint_path)
latest = tf.train.latest_checkpoint(checkpoint_dir)
model.load_weights(latest)

loss, acc = model.evaluate(x_test, y_test)

262/262 [==============================] - 0s 244us/sample - loss: 0.4393 - acc: 0.7977

你可能感兴趣的:(TensorFlow基础笔记)

数据仓库——维度表一致性墨染丶eye 背诵数据仓库
数据仓库基础笔记思维导图已经整理完毕，完整连接为：数据仓库基础知识笔记思维导图维度一致性问题从逻辑层面来看，当一系列星型模型共享一组公共维度时，所涉及的维度称为一致性维度。当维度表存在不一致时，短期的成功难以弥补长期的错误。维度时确保不同过程中信息集成起来实现横向钻取货活动的关键。造成横向钻取失败的原因维度结构的差别，因为维度的差别，分析工作涉及的领域从简单到复杂，但是都是通过复杂的报表来弥补设计
Python开发常用的三方模块如下：换个网名有点难 python 开发语言
Python是一门功能强大的编程语言，拥有丰富的第三方库，这些库为开发者提供了极大的便利。以下是100个常用的Python库，涵盖了多个领域：1、NumPy，用于科学计算的基础库。2、Pandas，提供数据结构和数据分析工具。3、Matplotlib，一个绘图库。4、Scikit-learn，机器学习库。5、SciPy，用于数学、科学和工程的库。6、TensorFlow，由Google开发的开源机
关于python版本与TensorFlow安装的版本问题 iiimharrygGc. python tensorflow 开发语言
实测在conda环境下，python3.12的版本无法安装TensorFlow2.14.0（截至2024.5.21）最新版本在python3.7版本下正常安装ps：上述安装均在anacondanavigator软件内安装
Vue + Django的人脸识别系统 DXSsssss python DRF tensorflow 人脸识别
最近在研究机器学习，刚好最近看了vue+Djangodrf的一些课程，学以致用，做了一个人脸识别系统。项目前端使用Vue框架，用到了elementui组件，写起来真是方便。比之前传统的dtl方便了太多。后端使用了drf，识别知识刚开始打算使用opencv+tensorflow,但是发现吧识别以后的结果返回到浏览器当中时使用opencv比较麻烦（主要是我太菜，想不到比较好的方法），因此最终使用了tf
Awesome TensorFlow weixin_30594001 人工智能移动开发大数据
AwesomeTensorFlowAcuratedlistofawesomeTensorFlowexperiments,libraries,andprojects.Inspiredbyawesome-machine-learning.WhatisTensorFlow?TensorFlowisanopensourcesoftwarelibraryfornumericalcomputationusin
R语言基础笔记 waterHBO r语言笔记开发语言
起因:今天不知道要写什么。把之前的笔记复制一下。代码开头，导入:#清除系统变量rm(list=ls())#隐藏警告信息:options(warn=-1)#把当前目录，设置为工作目录。library(rstudioapi)current_folder_path0.0&ideology<10.0)分组聚合，类似groupby()df2<-aggregate(df1KaTeXparseerror:Exp
【ShuQiHere】小白也能懂的 TensorFlow 和 PyTorch GPU 配置教程 ShuQiHere tensorflow pytorch 人工智能
【ShuQiHere】在深度学习中，GPU的使用对于加速模型训练至关重要。然而，对于许多刚刚入门的小白来说，如何在TensorFlow和PyTorch中指定使用GPU进行训练可能会感到困惑。在本文中，我将详细介绍如何在这两个主流的深度学习框架中指定使用GPU进行训练，并确保每一个步骤都简单易懂，跟着我的步骤来，你也能轻松上手！1.安装所需库首先，确保你已经安装了TensorFlow或PyTorch
TensorFlow的基本概念以及使用场景张柏慈决策树
TensorFlow是一个机器学习平台，用于构建和训练机器学习模型。它使用图形表示计算任务，其中节点表示数学操作，边表示计算之间的数据流动。TensorFlow的主要特点包括：1.多平台支持：TensorFlow可以运行在多种硬件和操作系统上，包括CPU、GPU和移动设备。2.自动求导：TensorFlow可以自动计算模型参数的梯度，通过优化算法更新参数，以提高模型的准确性。3.分布式计算：Ten
基于VGG的猫狗识别卑微小鹿 tensorflow tensorflow
由于猫和狗的数据在这里，所以就做了一下分类的神经网络1、首先进行图像处理：importcsvimportglobimportosimportrandomos.environ['TF_CPP_MIN_LOG_LEVEL']='2'importtensorflowastffromtensorflowimportkerasfromtensorflow.kerasimportlayersimportnum
轻松升级：Ollama + OpenWebUI 安装与配置【AIStarter】 ai_xiaogui AI作画 AI软件人工智能 AI写作 AIStarter
Ollama是一个开源项目，用于构建和训练大规模语言模型，而OpenWebUI则提供了一个方便的前端界面来管理和监控这些模型。本文将指导你如何更新这两个工具，并顺利完成配置。准备工作确保你的系统已安装Git和Python环境。安装必要的依赖库，如TensorFlow或PyTorch等。更新步骤克隆项目：使用Git命令行工具克隆最新的Ollama和OpenWebUI仓库到本地。更新代码：确保你正在使
深度学习之基于Tensorflow卷积神经网络水果蔬菜分类识别系统 qq1744828575 python python plotly
欢迎大家点赞、收藏、关注、评论啦，由于篇幅有限，只展示了部分核心代码。文章目录一项目简介二、功能三、系统四.总结一项目简介一、项目背景与目标背景：在现代农业、智能零售等领域，自动化分类与识别技术对于提高效率、优化供应链管理具有重要意义。为了响应这一需求，本项目旨在构建一个基于深度学习技术的水果蔬菜分类识别系统。目标：构建一个准确率高、性能稳定的水果蔬菜分类识别模型，利用Tensorflow框架
ImportError: cannot import name ‘conv_utils‘ from ‘keras.utils‘ CheCacao keras 深度学习 python tensorflow tensorflow2 人工智能
将fromkeras.utilsimportconv_utils改为fromtensorflow.python.keras.utilsimportconv_utilsImportError:nomodulenamed'tensorflow.keras.engine将fromkeras.engine.topologyimportLayer改为fromtensorflow.python.keras.l
jupyter出错ImportError: cannot import name ‘np_utils‘ from ‘keras.utils‘ ，怎么解决？七月初七淮水竹亭～人工智能 python jupyter keras 深度学习
文章前言此篇文章主要是记录一下我遇到的问题以及我是如何解决的，希望下次遇到类似问题可以很快解决。此外，也希望能帮助到大家。遇到的问题出错：ImportError:cannotimportname'np_utils'from'keras.utils'，如图：如何解决首先我根据网上文章的一些提示，将fromkeras.utilsimportnp_utils换成了fromtensorflow.keras
tensorflow和python不兼容_tensorflow与numpy的版本兼容性问题的解决 weixin_39727934
在Python交互式窗口导入tensorflow出现了下面的错误：ox8免费资源网root@ubuntu:~#python3Python3.6.8(default,Oct72019,12:59:55)[GCC8.3.0]onlinuxType"help","copyright","credits"or"license"formoreinformation.>>>importtensorflowas
numpy python 兼容_tensorflow与numpy的版本兼容性问题 weixin_39761822 numpy python 兼容
在Python交互式窗口导入tensorflow出现了下面的错误：root@ubuntu:~#python3Python3.6.8(default,Oct72019,12:59:55)[GCC8.3.0]onlinuxType"help","copyright","credits"or"license"formoreinformation.>>>importtensorflowastf;/usr/
安装tensorflow2.5.0 发现 tensorflow 和 numba 两者对应Numpy版本冲突 GJK_ tensorflow numpy 人工智能
问题：python3.8安装tensorflow2.5.0发现tensorflow和numba两者对应Numpy版本冲突tensorflow-gpu2.5.0requiresnumpy~=1.19.2numba0.58.1requiresnumpy=1.22解决方法：将numba降低版本为0.53pipinstallnumba==0.53再将numpy版本改为1.19.2pipinstallnum
python3.7安装keras教程_python 3.7 安装 sklearn keras(tf.keras) weixin_39641103
#1sklearn一般方法网上有很多教程，不再赘述。注意顺序是numpy+mkl，然后scipy的环境，scipy，然后sklearn#2anocondaanaconda原始的环境已经自带了sklearn，这里说一下新建环境（比如创建了一个tensorflow的环境），activatetensorflow2.0，然后condainstallsklearn即可，会帮你把各种需要的库都安装。#kera
Python高层神经网络 API库之Keras使用详解 Rocky006 python keras 开发语言
概要随着深度学习在各个领域的广泛应用，许多开发者开始使用各种框架来构建和训练神经网络模型。Keras是一个高层神经网络API，使用Python编写，并能够运行在TensorFlow、CNTK和Theano之上。Keras旨在简化深度学习模型的构建过程，使得开发者能够更加专注于实验和研究。本文将详细介绍Keras库，包括其安装方法、主要特性、基本和高级功能，以及实际应用场景，帮助全面了解并掌握该库的
将本地已有的项目上传到新建的git仓库的方法 10676
将本地已有的一个非git项目上传到新建的git仓库的方法一共有两种。一、克隆+拷贝第一种方法比较简单，直接用把远程仓库拉到本地，然后再把自己本地的项目拷贝到仓库中去。然后push到远程仓库上去即可。此方法适用于本地项目不是一个git仓库的情况。1、首先克隆[email protected]:yuanmingchen/tensorflow_study.git2、然后复制自己项目的所有文件到
Python(PyTorch和TensorFlow)图像分割卷积网络导图(生物医学) 亚图跨际交叉知识 Python 生物医学脑肿瘤图像皮肤病变多模态医学图像多尺度特征生物医学腹部胰腺图像病灶边界气胸图像
要点语义分割图像三层分割椭圆图像脑肿瘤图像分割动物图像分割皮肤病变分割多模态医学图像多尺度特征生物医学肖像多类和医学分割通用图像分割模板腹部胰腺图像分割分类注意力网络病灶边界分割气胸图像分割Python生物医学图像卷积网络该网络由收缩路径和扩展路径组成，收缩路径是一种典型的卷积网络，由重复应用卷积组成，每个卷积后跟一个整流线性单元(ReLU)和一个最大池化操作。在收缩过程中，空间信息减少，而特征信
一口气了解大模型相关通识，基础笔记！ AI小白熊笔记数据库架构面试职场和发展 transformer ai
一、大模型生态有哪些语言类大模型:GPT-3、GPT-3.5、GPT-4系列模型。并且，OpenAl在训练GPT-3的同时训练了参数不同、复杂度各不相同的A、B、C、D四项大模型(基座模型)，用于不同场景的应用;其中，A、B、C、D模型的全称分别是ada、babbage、curie(居里)和davinci（达芬奇），四个模型并不是GPT-3的微调模型，而是独立训练的四个模型;四个模型的参数规模和复
AttributeError: ‘tuple‘ object has no attribute ‘shape‘ 晓胡同学 keras 深度学习 tensorflow
AttributeError:‘tuple’objecthasnoattribute‘shape’在将keras代码改为tensorflow2代码的时候报了如下错误AttributeError:'tuple'objecthasnoattribute'shape'经过调查发现，损失函数写错了原来的是这样model.compile(loss=['binary_crossentropy'],optimi
Autoencoder chuange6363 人工智能 python
自编码器Autoencoder稀疏自编码器SparseAutoencoder降噪自编码器DenoisingAutoencoder堆叠自编码器StackedAutoencoder本博客是从梁斌博士的博客上面复制过来的，本人利用Tensorflow重新实现了博客中的代码深度学习有一个重要的概念叫autoencoder，这是个什么东西呢，本文通过一个例子来普及这个术语。简单来说autoencoder是一
OSError: libnccl.so.2: cannot open shared object file: No such file or directory 王小葱鸭 python
linux安装完torch或者tensorflow的gpu版本，安装没问题，但是import就有问题，报错OSError:libnccl.so.2:cannotopensharedobjectfile:Nosuchfileordirectory，是缺少nvidia的ncll，下面介绍解决方法：1安装ncll下载链接https://developer.nvidia.com/nccl/nccl-dow
深度学习框架人工智能操作系统训练&前向推理 PyTorch Tensorflow MindSpore caffe 张量加速引擎TBE 深度学习编译器多面体 polyhedral AI集群框架 EwenWanW 深度学习人工智能 pytorch 深度学习编译器
深度学习框架人工智能操作系统训练&前向推理深度学习框架发展到今天，目前在架构上大体已经基本上成熟并且逐渐趋同。无论是国外的Tensorflow、PyTorch，亦或是国内最近开源的MegEngine、MindSpore，目前基本上都是支持EagerMode和GraphMode两种模式。AI嵌入式框架OneFlow&清华计图Jittor&华为深度学习框架MindSpore&旷视深度学习框架MegEn
Tensorflow中Keras搭建神经网络六步法及参数详解 -- Tensorflow自学笔记12 青瓷看世界 tensorflow 笔记人工智能深度学习神经网络
一.tf.keras搭建神经网络六步法1.import相关模块如importtensorflowastf。2.指定输入网络的训练集和测试集如指定训练集的输入x_train和标签y_train，测试集的输入x_test和标签y_test。3.逐层搭建网络结构model=tf.keras.models.Sequential()。4.在model.compile()中配置训练方法选择训练时使用的优化器、
MacBook Air M1 使用 miniconda 安装python3.11.7 和 tensorflow2.16.1详细 - TensorFlow自学笔记1 青瓷看世界 tensorflow tensorflow 人工智能 mac python 深度学习
1m1mac安装xcode命令工具在Terminal终端执行以下代码：xcode-select--install2下载支持m1芯片arm64的miniconda在miniconda官网，找到下图中保护AppleM1的bash安装包，Miniconda—Anacondadocumentation3安装miniconda在Terminal执行下列代码：1）cd"miniconda下载目录"2）bash
机器学习框架巅峰对决：TensorFlow vs. PyTorch vs. Scikit-Learn实战分析 @sinner 技术选型机器学习 tensorflow pytorch scikit-learn
1.引言1.1机器学习框架的重要性在机器学习的黄金时代，框架的选择对于开发高效、可扩展的模型至关重要。合适的框架可以极大地提高开发效率，简化模型的构建和训练过程，并支持大规模的模型部署。因此，了解和选择最合适的机器学习框架对于研究人员和工程师来说是一个关键的步骤。1.2三大框架概览：TensorFlow、PyTorch、Scikit-Learn目前，最流行的机器学习框架主要有TensorFlow、
Tensorflow2.16中的Keras包含哪些层(layers)?层的功能及参数详细解释 -- TensorFlow自学笔记6 青瓷看世界 tensorflow tensorflow keras 人工智能
在Keras中，层（Layer）是构建神经网络的基本组件。Keras提供了多种类型的层，用于处理不同类型的输入数据和执行特定的数学操作。英文版可参考TensorFlow官方文档：Module:tf.keras.layers|TensorFlowv2.16.1一.层的分类‌1.基本网络层‌1.1.Dense层用于执行全连接操作；1.2.卷积层Conv1D、Conv2D、Conv3D层，用于执行一维、
Tensorflow2 如何扩展现有数据集(缩放、随机旋转、水平翻转、平移等)，从而提高模型的准确率 -- Tensorflow自学笔记14 青瓷看世界 tensorflow 人工智能 python
实际生活中的数据集，往往不是标准的数据，而是有倾斜角度、有旋转、有偏移的数据，为了提高数据集的真实性，提高模型预测的准确率，可以用ImageDataGenerator函数来扩展数据集importtensorflowastffromtensorflow.keras.preprocessing.imageimportImageDataGeneratorimage_gen_train=ImageData
深入浅出Java Annotation(元注解和自定义注解） Josh_Persistence Java Annotation 元注解自定义注解
一、基本概述　　 Annontation是Java5开始引入的新特征。中文名称一般叫注解。它提供了一种安全的类似注释的机制，用来将任何的信息或元数据（metadata）与程序元素（类、方法、成员变量等）进行关联。　　更通俗的意思是为程序的元素（类、方法、成员变量）加上更直观更明了的说明，这些说明信息是与程序的业务逻辑无关，并且是供指定的工具或
mysql优化特定类型的查询 annan211 java 工作 mysql
本节所介绍的查询优化的技巧都是和特定版本相关的，所以对于未来mysql的版本未必适用。 1 优化count查询对于count这个函数的网上的大部分资料都是错误的或者是理解的都是一知半解的。在做优化之前我们先来看看真正的count()函数的作用到底是什么。 count()是一个特殊的函数，有两种非常不同的作用，他可以统计某个列值的数量，也可以统计行数。在统
MAC下安装多版本JDK和切换几种方式棋子chessman jdk
环境： MAC AIR,OS X 10.10,64位历史：过去 Mac 上的 Java 都是由 Apple 自己提供，只支持到 Java 6，并且OS X 10.7 开始系统并不自带（而是可选安装）（原自带的是1.6）。后来 Apple 加入 OpenJDK 继续支持 Java 6，而 Java 7 将由 Oracle 负责提供。在终端中输入jav
javaScript （1） Array_06 JavaScript java 浏览器
JavaScript 1、运算符　　运算符就是完成操作的一系列符号，它有七类：　　赋值运算符（=,+=,-=,*=,/=,%=,<<=,>>=,|=,&=）、算术运算符(+,-,*,/,++,--,%)、比较运算符(>,<,<=,>=,==,===,!=,!==)、逻辑运算符(||,&&,!)、条件运算(?:)、位
国内顶级代码分享网站袁潇含 java jdk oracle .net PHP
现在国内很多开源网站感觉都是为了利益而做的当然利益是肯定的,否则谁也不会免费的去做网站 &
Elasticsearch、MongoDB和Hadoop比较随意而生 mongodb hadoop 搜索引擎
IT界在过去几年中出现了一个有趣的现象。很多新的技术出现并立即拥抱了“大数据”。稍微老一点的技术也会将大数据添进自己的特性，避免落大部队太远，我们看到了不同技术之间的边际的模糊化。假如你有诸如Elasticsearch或者Solr这样的搜索引擎，它们存储着JSON文档，MongoDB存着JSON文档，或者一堆JSON文档存放在一个Hadoop集群的HDFS中。你可以使用这三种配
mac os 系统科研软件总结张亚雄 mac os
1.1 Microsoft Office for Mac 2011 大客户版，自行搜索。 1.2 Latex （MacTex）: 系统环境：https://tug.org/mactex/ &nb
Maven实战（四）生命周期 AdyZhang maven
1. 三套生命周期 Maven拥有三套相互独立的生命周期，它们分别为clean，default和site。每个生命周期包含一些阶段，这些阶段是有顺序的，并且后面的阶段依赖于前面的阶段，用户和Maven最直接的交互方式就是调用这些生命周期阶段。以clean生命周期为例，它包含的阶段有pre-clean, clean 和 post
Linux下Jenkins迁移 aijuans Jenkins
1. 将Jenkins程序目录copy过去源程序在/export/data/tomcatRoot/ofctest-jenkins.jd.com下面 tar -cvzf jenkins.tar.gz ofctest-jenkins.jd.com &
request.getInputStream()只能获取一次的问题 ayaoxinchao request Inputstream
问题：在使用HTTP协议实现应用间接口通信时，服务端读取客户端请求过来的数据，会用到request.getInputStream()，第一次读取的时候可以读取到数据，但是接下来的读取操作都读取不到数据原因： 1. 一个InputStream对象在被读取完成后，将无法被再次读取，始终返回-1； 2. InputStream并没有实现reset方法（可以重
数据库SQL优化大总结之百万级数据库优化方案 BigBird2012 SQL优化
网上关于SQL优化的教程很多，但是比较杂乱。近日有空整理了一下，写出来跟大家分享一下，其中有错误和不足的地方，还请大家纠正补充。这篇文章我花费了大量的时间查找资料、修改、排版，希望大家阅读之后，感觉好的话推荐给更多的人，让更多的人看到、纠正以及补充。 1.对查询进行优化，要尽量避免全表扫描，首先应考虑在 where 及 order by 涉及的列上建立索引。 2.应尽量避免在 where
jsonObject的使用 bijian1013 java json
在项目中难免会用java处理json格式的数据，因此封装了一个JSONUtil工具类。 JSONUtil.java package com.bijian.json.study; import java.util.ArrayList; import java.util.Date; import java.util.HashMap;
[Zookeeper学习笔记之六]Zookeeper源代码分析之Zookeeper.WatchRegistration bit1129 zookeeper
Zookeeper类是Zookeeper提供给用户访问Zookeeper service的主要API，它包含了如下几个内部类首先分析它的内部类，从WatchRegistration开始，为指定的znode path注册一个Watcher， /** * Register a watcher for a particular p
【Scala十三】Scala核心七：部分应用函数 bit1129 scala
何为部分应用函数？ Partially applied function: A function that’s used in an expression and that misses some of its arguments.For instance, if function f has type Int => Int => Int, then f and f(1) are p
Tomcat Error listenerStart 终极大法 ronin47 tomcat
Tomcat报的错太含糊了，什么错都没报出来，只提示了Error listenerStart。为了调试，我们要获得更详细的日志。可以在WEB-INF/classes目录下新建一个文件叫logging.properties，内容如下 Java代码 handlers = org.apache.juli.FileHandler, java.util.logging.ConsoleHa
不用加减符号实现加减法 BrokenDreams 实现
今天有群友发了一个问题，要求不用加减符号(包括负号)来实现加减法。分析一下，先看最简单的情况，假设1+1，按二进制算的话结果是10，可以看到从右往左的第一位变为0，第二位由于进位变为1。
读《研磨设计模式》-代码笔记-状态模式-State bylijinnan java 设计模式
声明：本文只为方便我个人查阅和理解，详细的分析以及源代码请移步原作者的博客http://chjavach.iteye.com/ /* 当一个对象的内在状态改变时允许改变其行为，这个对象看起来像是改变了其类状态模式主要解决的是当控制一个对象状态的条件表达式过于复杂时的情况把状态的判断逻辑转移到表示不同状态的一系列类中，可以把复杂的判断逻辑简化如果在
CUDA程序block和thread超出硬件允许值时的异常 cherishLC CUDA
调用CUDA的核函数时指定block 和 thread大小，该大小可以是dim3类型的（三维数组），只用一维时可以是usigned int型的。以下程序验证了当block或thread大小超出硬件允许值时会产生异常！！！GPU根本不会执行运算！！！所以验证结果的正确性很重要！！！在VS中创建CUDA项目会有一个模板，里面有更详细的状态验证。以下程序在K5000GPU上跑的。
诡异的超长时间GC问题定位 chenchao051 jvm cms GC hbase swap
HBase的GC策略采用PawNew+CMS, 这是大众化的配置，ParNew经常会出现停顿时间特别长的情况，有时候甚至长到令人发指的地步，例如请看如下日志： 2012-10-17T05:54:54.293+0800: 739594.224: [GC 739606.508: [ParNew: 996800K->110720K(996800K), 178.8826900 secs] 3700
maven环境快速搭建 daizj 安装 mavne 环境配置
一下载maven 安装maven之前，要先安装jdk及配置JAVA_HOME环境变量。这个安装和配置java环境不用多说。 maven下载地址：http://maven.apache.org/download.html，目前最新的是这个apache-maven-3.2.5-bin.zip，然后解压在任意位置，最好地址中不要带中文字符，这个做java 的都知道，地址中出现中文会出现很多
PHP网站安全，避免PHP网站受到攻击的方法 dcj3sjt126com PHP
对于PHP网站安全主要存在这样几种攻击方式:1、命令注入(Command Injection)2、eval注入(Eval Injection)3、客户端脚本攻击(Script Insertion)4、跨网站脚本攻击(Cross Site Scripting, XSS)5、SQL注入攻击(SQL injection)6、跨网站请求伪造攻击(Cross Site Request Forgerie
yii中给CGridView设置默认的排序根据时间倒序的方法 dcj3sjt126com GridView
public function searchWithRelated() { $criteria = new CDbCriteria; $criteria->together = true; //without th
Java集合对象和数组对象的转换 dyy_gusi java集合
在开发中，我们经常需要将集合对象（List，Set）转换为数组对象，或者将数组对象转换为集合对象。Java提供了相互转换的工具，但是我们使用的时候需要注意，不能乱用滥用。 1、数组对象转换为集合对象最暴力的方式是new一个集合对象，然后遍历数组，依次将数组中的元素放入到新的集合中，但是这样做显然过
nginx同一主机部署多个应用 geeksun nginx
近日有一需求，需要在一台主机上用nginx部署2个php应用，分别是wordpress和wiki，探索了半天，终于部署好了，下面把过程记录下来。 1. 在nginx下创建vhosts目录，用以放置vhost文件。 mkdir vhosts 2. 修改nginx.conf的配置，在http节点增加下面内容设置，用来包含vhosts里的配置文件 #
ubuntu添加admin权限的用户账号 hongtoushizi ubuntu useradd
ubuntu创建账号的方式通常用到两种：useradd 和adduser . 本人尝试了useradd方法，步骤如下： 1:useradd 使用useradd时，如果后面不加任何参数的话，如：sudo useradd sysadm 创建出来的用户将是默认的三无用户：无home directory ,无密码,无系统shell。顾应该如下操作：
第五章常用Lua开发库2-JSON库、编码转换、字符串处理 jinnianshilongnian nginx lua
JSON库在进行数据传输时JSON格式目前应用广泛，因此从Lua对象与JSON字符串之间相互转换是一个非常常见的功能；目前Lua也有几个JSON库，本人用过cjson、dkjson。其中cjson的语法严格（比如unicode \u0020\u7eaf），要求符合规范否则会解析失败（如\u002），而dkjson相对宽松，当然也可以通过修改cjson的源码来完成
Spring定时器配置的两种实现方式OpenSymphony Quartz和java Timer详解 yaerfeng1989 timer quartz 定时器
原创整理不易，转载请注明出处：Spring定时器配置的两种实现方式OpenSymphony Quartz和java Timer详解代码下载地址：http://www.zuidaima.com/share/1772648445103104.htm 有两种流行Spring定时器配置：Java的Timer类和OpenSymphony的Quartz。 1.Java Timer定时首先继承jav
Linux下df与du两个命令的差别？ pda158 linux
　一、df显示文件系统的使用情况，与du比較，就是更全盘化。　　最经常使用的就是 df -T，显示文件系统的使用情况并显示文件系统的类型。　　举比例如以下：　　[root@localhost ~]# df -T 　　Filesystem Type &n
[转]SQLite的工具类 ---- 通过反射把Cursor封装到VO对象 ctfzh VO android sqlite 反射 Cursor
在写DAO层时，觉得从Cursor里一个一个的取出字段值再装到VO(值对象)里太麻烦了，就写了一个工具类，用到了反射，可以把查询记录的值装到对应的VO里，也可以生成该VO的List。使用时需要注意：考虑到Android的性能问题，VO没有使用Setter和Getter，而是直接用public的属性。表中的字段名需要和VO的属性名一样，要是不一样就得在查询的SQL中
该学习笔记用到的Employee表 vipbooks oracle sql 工作
这是我在学习Oracle是用到的Employee表，在该笔记中用到的就是这张表，大家可以用它来学习和练习。 drop table Employee; -- 员工信息表 create table Employee( -- 员工编号 EmpNo number(3) primary key, -- 姓