Keras用多层感知机探索Titanic数据

GITHUB地址https://github.com/fz861062923/Keras

数据来源:http://biostat.mc.vanderbilt.edu/wiki/Main/DataSets

查看数据

import numpy as np
import pandas as pd

df=pd.read_excel('titanic3.xls')

df[:3]

	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

去除与survived无关的变量(eg.ticket,cabin)

cols=['survived','name','pclass','sex','age','sibsp','parch','fare','embarked']
last_df=df[cols]

last_df[:2]

	survived	name	pclass	sex	age	sibsp	parch	fare	embarked
0	1	Allen, Miss. Elisabeth Walton	1	female	29.0000	0	0	211.3375	S
1	1	Allison, Master. Hudson Trevor	1	male	0.9167	1	2	151.5500	S

数据预处理

#先drop掉name,目前还用不到
last_df=last_df.drop(['name'],axis=1)

last_df[:2]

	survived	pclass	sex	age	sibsp	parch	fare	embarked
0	1	1	female	29.0000	0	0	211.3375	S
1	1	1	male	0.9167	1	2	151.5500	S

last_df.describe()

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

last_df.dtypes

survived      int64
pclass        int64
sex          object
age         float64
sibsp         int64
parch         int64
fare        float64
embarked     object
dtype: object

#接下来将age,fare里面的NULL全部用平均值来代替
age_mean=last_df['age'].mean()
last_df['age']=last_df['age'].fillna(age_mean)
fare_mean=last_df['fare'].mean()
last_df['fare']=last_df['fare'].fillna(fare_mean)

#最后一个问题将sex变为数值型变量，便于分类，使用map函数方便快捷
last_df['sex']=last_df['sex'].map({'female':0,'male':1})

last_df

	survived	pclass	sex	age	sibsp	parch	fare	embarked
0	1	1	0	29.000000	0	0	211.3375	S
1	1	1	1	0.916700	1	2	151.5500	S
2	0	1	0	2.000000	1	2	151.5500	S
3	0	1	1	30.000000	1	2	151.5500	S
4	0	1	0	25.000000	1	2	151.5500	S
5	1	1	1	48.000000	0	0	26.5500	S
6	1	1	0	63.000000	1	0	77.9583	S
7	0	1	1	39.000000	0	0	0.0000	S
8	1	1	0	53.000000	2	0	51.4792	S
9	0	1	1	71.000000	0	0	49.5042	C
10	0	1	1	47.000000	1	0	227.5250	C
11	1	1	0	18.000000	1	0	227.5250	C
12	1	1	0	24.000000	0	0	69.3000	C
13	1	1	0	26.000000	0	0	78.8500	S
14	1	1	1	80.000000	0	0	30.0000	S
15	0	1	1	29.881135	0	0	25.9250	S
16	0	1	1	24.000000	0	1	247.5208	C
17	1	1	0	50.000000	0	1	247.5208	C
18	1	1	0	32.000000	0	0	76.2917	C
19	0	1	1	36.000000	0	0	75.2417	C
20	1	1	1	37.000000	1	1	52.5542	S
21	1	1	0	47.000000	1	1	52.5542	S
22	1	1	1	26.000000	0	0	30.0000	C
23	1	1	0	42.000000	0	0	227.5250	C
24	1	1	0	29.000000	0	0	221.7792	S
25	0	1	1	25.000000	0	0	26.0000	C
26	1	1	1	25.000000	1	0	91.0792	C
27	1	1	0	19.000000	1	0	91.0792	C
28	1	1	0	35.000000	0	0	135.6333	S
29	1	1	1	28.000000	0	0	26.5500	S
...	...	...	...	...	...	...	...	...
1279	0	3	0	14.000000	0	0	7.8542	S
1280	0	3	1	22.000000	0	0	7.8958	S
1281	0	3	1	22.000000	0	0	9.0000	S
1282	0	3	1	29.881135	0	0	8.0500	S
1283	0	3	1	29.881135	0	0	7.5500	S
1284	0	3	1	29.881135	0	0	8.0500	S
1285	0	3	1	32.500000	0	0	9.5000	S
1286	1	3	0	38.000000	0	0	7.2292	C
1287	0	3	1	51.000000	0	0	7.7500	S
1288	0	3	1	18.000000	1	0	6.4958	S
1289	0	3	1	21.000000	1	0	6.4958	S
1290	1	3	0	47.000000	1	0	7.0000	S
1291	0	3	1	29.881135	0	0	8.7125	S
1292	0	3	1	29.881135	0	0	7.5500	S
1293	0	3	1	29.881135	0	0	8.0500	S
1294	0	3	1	28.500000	0	0	16.1000	S
1295	0	3	1	21.000000	0	0	7.2500	S
1296	0	3	1	27.000000	0	0	8.6625	S
1297	0	3	1	29.881135	0	0	7.2500	S
1298	0	3	1	36.000000	0	0	9.5000	S
1299	0	3	1	27.000000	1	0	14.4542	C
1300	1	3	0	15.000000	1	0	14.4542	C
1301	0	3	1	45.500000	0	0	7.2250	C
1302	0	3	1	29.881135	0	0	7.2250	C
1303	0	3	1	29.881135	0	0	14.4583	C
1304	0	3	0	14.500000	1	0	14.4542	C
1305	0	3	0	29.881135	1	0	14.4542	C
1306	0	3	1	26.500000	0	0	7.2250	C
1307	0	3	1	27.000000	0	0	7.2250	C
1308	0	3	1	29.000000	0	0	7.8750	S

1309 rows × 8 columns

#对登船港口进行一位有效位编码
last2_df=pd.get_dummies(data=last_df,columns=['embarked'])#这个方法真的很人性化

last2_df[:2]#查看预处理后的结果

	survived	pclass	sex	age	sibsp	parch	fare	embarked_C	embarked_Q	embarked_S
0	1	1	0	29.0000	0	0	211.3375	0	0	1
1	1	1	1	0.9167	1	2	151.5500	0	0	1

DF转换为array

array=last2_df.values

array.shape

(1309, 10)

array[:1]

array([[  1.    ,   1.    ,   0.    ,  29.    ,   0.    ,   0.    ,
        211.3375,   0.    ,   0.    ,   1.    ]])

第一个字段为label,后面的特征全部可以定义为feature

label=array[:,0]#可以说这是一个二维的向量，只需要每一行的第一列即可
feature=array[:,1:]

label

array([1., 1., 0., ..., 0., 0., 0.])

feature

array([[ 1.    ,  0.    , 29.    , ...,  0.    ,  0.    ,  1.    ],
       [ 1.    ,  1.    ,  0.9167, ...,  0.    ,  0.    ,  1.    ],
       [ 1.    ,  0.    ,  2.    , ...,  0.    ,  0.    ,  1.    ],
       ...,
       [ 3.    ,  1.    , 26.5   , ...,  1.    ,  0.    ,  0.    ],
       [ 3.    ,  1.    , 27.    , ...,  1.    ,  0.    ,  0.    ],
       [ 3.    ,  1.    , 29.    , ...,  0.    ,  0.    ,  1.    ]])

这里有个问题年龄，运费fare，sex之间的数值差异十分大

特征字段进行标准化

from sklearn import preprocessing#sklearn的processing模块有具体介绍
scale=preprocessing.MinMaxScaler(feature_range=(0,1))
last_feature=scale.fit_transform(feature)

last_feature

array([[0.        , 0.        , 0.36116884, ..., 0.        , 0.        ,
        1.        ],
       [0.        , 1.        , 0.00939458, ..., 0.        , 0.        ,
        1.        ],
       [0.        , 0.        , 0.0229641 , ..., 0.        , 0.        ,
        1.        ],
       ...,
       [1.        , 1.        , 0.32985358, ..., 1.        , 0.        ,
        0.        ],
       [1.        , 1.        , 0.33611663, ..., 1.        , 0.        ,
        0.        ],
       [1.        , 1.        , 0.36116884, ..., 0.        , 0.        ,
        1.        ]])

金星：完美_·

将数据分为train和test两部分

cols=['survived','name','pclass','sex','age','sibsp','parch','fare','embarked']
_df=df[cols]
msk=np.random.rand(len(_df))<0.8
train=_df[msk]
test=_df[~msk]

len(train)

1071

len(test)

238

len(df)

1309

len(df)==len(train)+len(test)

True

将上面预处理数据的过程写成一个完整的函数

from sklearn import preprocessing
def preprocess(df):
    df=df.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})
    df=pd.get_dummies(data=df,columns=['embarked'])
    array=df.values
    Label=array[:,0]
    feature=array[:,1:]
    scale=preprocessing.MinMaxScaler(feature_range=(0,1))
    Feature=scale.fit_transform(feature)
    return Feature,Label

train_feature,train_label=preprocess(train)
test_feature,test_label=preprocess(test)

train_feature

array([[0.        , 1.        , 0.01015802, ..., 0.        , 0.        ,
        1.        ],
       [0.        , 0.        , 0.02483026, ..., 0.        , 0.        ,
        1.        ],
       [0.        , 1.        , 0.40406294, ..., 0.        , 0.        ,
        1.        ],
       ...,
       [1.        , 0.        , 0.39950884, ..., 1.        , 0.        ,
        0.        ],
       [1.        , 1.        , 0.35665885, ..., 1.        , 0.        ,
        0.        ],
       [1.        , 1.        , 0.39051891, ..., 0.        , 0.        ,
        1.        ]])

train_label

array([1., 0., 0., ..., 0., 0., 0.])

建立多层感知机模型

from keras.models import Sequential
from keras.layers import Dense,Dropout

C:\Users\admin\AppData\Local\conda\conda\envs\tensorflow\lib\site-packages\h5py\__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.
  from ._conv import register_converters as _register_converters
Using TensorFlow backend.

model=Sequential()

建立隐藏层1

model.add(Dense(units=40,input_dim=9,#设置40个神经元
                kernel_initializer='uniform',#使用uniform分布的随机数，初始化权重和bias
               activation='relu'))

建立隐藏层2

model.add(Dense(units=30,kernel_initializer='uniform',
               activation='relu'))

建立输出层

model.add(Dense(units=1,kernel_initializer='uniform',#输出层一个神经元
               activation='sigmoid'))

开始训练

model.compile(loss='binary_crossentropy',
             optimizer='adam',metrics=['accuracy'])

train_history=model.fit(train_feature,train_label,validation_split=0.1,
                  epochs=30,
                  batch_size=30,
                  verbose=2)

Train on 963 samples, validate on 108 samples
Epoch 1/30
 - 1s - loss: 0.6886 - acc: 0.5971 - val_loss: 0.6607 - val_acc: 0.7963
Epoch 2/30
 - 0s - loss: 0.6632 - acc: 0.6199 - val_loss: 0.5776 - val_acc: 0.7963
Epoch 3/30
 - 0s - loss: 0.6057 - acc: 0.6615 - val_loss: 0.4881 - val_acc: 0.8148
Epoch 4/30
 - 0s - loss: 0.5522 - acc: 0.7549 - val_loss: 0.4740 - val_acc: 0.7685
Epoch 5/30
 - 0s - loss: 0.5153 - acc: 0.7695 - val_loss: 0.4613 - val_acc: 0.7778
Epoch 6/30
 - 0s - loss: 0.5006 - acc: 0.7726 - val_loss: 0.4621 - val_acc: 0.7778
Epoch 7/30
 - 0s - loss: 0.4930 - acc: 0.7726 - val_loss: 0.4668 - val_acc: 0.7778
Epoch 8/30
 - 0s - loss: 0.4873 - acc: 0.7705 - val_loss: 0.4570 - val_acc: 0.7778
Epoch 9/30
 - 0s - loss: 0.4816 - acc: 0.7767 - val_loss: 0.4726 - val_acc: 0.7778
Epoch 10/30
 - 0s - loss: 0.4805 - acc: 0.7757 - val_loss: 0.4551 - val_acc: 0.7778
Epoch 11/30
 - 0s - loss: 0.4751 - acc: 0.7767 - val_loss: 0.4482 - val_acc: 0.7778
Epoch 12/30
 - 0s - loss: 0.4728 - acc: 0.7819 - val_loss: 0.4719 - val_acc: 0.7778
Epoch 13/30
 - 0s - loss: 0.4733 - acc: 0.7767 - val_loss: 0.4423 - val_acc: 0.7963
Epoch 14/30
 - 0s - loss: 0.4688 - acc: 0.7747 - val_loss: 0.4606 - val_acc: 0.7778
Epoch 15/30
 - 0s - loss: 0.4743 - acc: 0.7653 - val_loss: 0.4357 - val_acc: 0.8241
Epoch 16/30
 - 0s - loss: 0.4660 - acc: 0.7882 - val_loss: 0.4416 - val_acc: 0.7963
Epoch 17/30
 - 0s - loss: 0.4648 - acc: 0.7799 - val_loss: 0.4356 - val_acc: 0.8241
Epoch 18/30
 - 0s - loss: 0.4641 - acc: 0.7850 - val_loss: 0.4331 - val_acc: 0.8148
Epoch 19/30
 - 0s - loss: 0.4648 - acc: 0.7934 - val_loss: 0.4429 - val_acc: 0.8056
Epoch 20/30
 - 0s - loss: 0.4608 - acc: 0.7830 - val_loss: 0.4356 - val_acc: 0.8241
Epoch 21/30
 - 0s - loss: 0.4640 - acc: 0.7902 - val_loss: 0.4362 - val_acc: 0.8241
Epoch 22/30
 - 0s - loss: 0.4579 - acc: 0.7923 - val_loss: 0.4422 - val_acc: 0.8148
Epoch 23/30
 - 0s - loss: 0.4587 - acc: 0.7882 - val_loss: 0.4333 - val_acc: 0.8241
Epoch 24/30
 - 0s - loss: 0.4573 - acc: 0.7861 - val_loss: 0.4327 - val_acc: 0.8241
Epoch 25/30
 - 0s - loss: 0.4559 - acc: 0.7871 - val_loss: 0.4326 - val_acc: 0.8333
Epoch 26/30
 - 0s - loss: 0.4549 - acc: 0.7913 - val_loss: 0.4368 - val_acc: 0.8148
Epoch 27/30
 - 0s - loss: 0.4544 - acc: 0.7913 - val_loss: 0.4310 - val_acc: 0.8241
Epoch 28/30
 - 0s - loss: 0.4559 - acc: 0.7850 - val_loss: 0.4299 - val_acc: 0.8241
Epoch 29/30
 - 0s - loss: 0.4533 - acc: 0.7923 - val_loss: 0.4324 - val_acc: 0.8241
Epoch 30/30
 - 1s - loss: 0.4544 - acc: 0.7944 - val_loss: 0.4299 - val_acc: 0.8333

可视化

import matplotlib.pyplot as plt
def show_train_history(train_history,train,validation):#输入参数分别为，train_history,
                                                        #训练数据的执行结果，验证数据的执行结果
    plt.plot(train_history.history[train])
    plt.plot(train_history.history[validation])
    plt.title('Train History')
    plt.ylabel(train)
    plt.xlabel('Epoch')
    plt.legend(['train', 'validation'], loc='upper left')
    plt.show()

show_train_history(train_history,'acc','val_acc')

show_train_history(train_history,'loss','val_loss')

评估模型准确率

scores =model.evaluate(test_feature,test_label)

238/238 [==============================] - 0s 67us/step

scores[1]

0.8319327736101231

做一个有趣的实验

模拟出Jack和Rose两个电影人物，比较他们的生存率

创建数据

Jack=pd.Series([0,'Jack',3,'male',23,1,0,5.000,'S'])
Rose=pd.Series([1,'Rose',1,'female',20,1,0,100.0000,'S'])
new_df=pd.DataFrame([list(Jack),list(Rose)],
                   columns=['survived','name','pclass','sex','age','sibsp','parch','fare','embarked'])

加入数据到DF中

df=pd.concat([_df,new_df],sort=False)

df[-4:]

	survived	name	pclass	sex	age	sibsp	parch	fare	embarked
1307	0	Zakarian, Mr. Ortin	3	male	27.0	0	0	7.225	C
1308	0	Zimmerman, Mr. Leo	3	male	29.0	0	0	7.875	S
0	0	Jack	3	male	23.0	1	0	5.000	S
1	1	Rose	1	female	20.0	1	0	100.000	S

进行预测

Feature,Label=preprocess(df)

probability=model.predict(Feature)

probability[:10]

array([[0.9776822 ],
       [0.54482657],
       [0.9709371 ],
       [0.36197898],
       [0.969016  ],
       [0.2566388 ],
       [0.9287646 ],
       [0.2852625 ],
       [0.9162269 ],
       [0.27310023]], dtype=float32)

pd=df
pd.insert(len(df.columns),'probability',probability)

pd[-2:]

	survived	name	pclass	sex	age	sibsp	parch	fare	embarked	probability
0	0	Jack	3	male	23.0	1	0	5.0	S	0.132897
1	1	Rose	1	female	20.0	1	0	100.0	S	0.964962

可以看出Rose的存活率的确远远大于Jack,滑稽xiao

• 看看生存概率大于90%却仍然挂了的数据

pd[(pd['survived']==0)& (pd['probability']>0.9)]

	survived	name	pclass	sex	age	sibsp	parch	fare	embarked	probability
2	0	Allison, Miss. Helen Loraine	1	female	2.0	1	2	151.5500	S	0.970937
4	0	Allison, Mrs. Hudson J C (Bessie Waldo Daniels)	1	female	25.0	1	2	151.5500	S	0.969016
105	0	Evans, Miss. Edith Corse	1	female	36.0	0	0	31.6792	C	0.970464
169	0	Isham, Miss. Ann Elizabeth	1	female	50.0	0	0	28.7125	C	0.967463
286	0	Straus, Mrs. Isidor (Rosalie Ida Blun)	1	female	63.0	1	0	221.7792	S	0.959756