딥러닝 모델 구현해 보기¶

학습 내용¶

  • 타이타닉 데이터 셋을 활용한 딥러닝 모델 구현해 보기
  • 첫번째 데이터 셋 : 자전거 공유 업체 시간대별 데이터
  • 두번째 데이터 셋 : 타이타닉 데이터 셋

목차

01. 라이브러리 및 데이터 불러오기
02. 입력 및 출력 지정
03. 딥러닝 구축 및 학습시키기

01. 라이브러리 및 데이터 불러오기

목차로 이동하기

In [1]:
import numpy as np
import matplotlib.pyplot as plt
import matplotlib
import pandas as pd
import tensorflow as tf
In [2]:
import keras
from keras.models import Sequential
from keras.layers import Dense
In [3]:
print(keras.__version__)

2.9.0
In [4]:
train = pd.read_csv("./titanic/train.csv")
test = pd.read_csv("./titanic/test.csv")
print(train.shape, test.shape)

(891, 12) (418, 11)
In [5]:
train.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 891 entries, 0 to 890
Data columns (total 12 columns):
 #   Column       Non-Null Count  Dtype  
---  ------       --------------  -----  
 0   PassengerId  891 non-null    int64  
 1   Survived     891 non-null    int64  
 2   Pclass       891 non-null    int64  
 3   Name         891 non-null    object 
 4   Sex          891 non-null    object 
 5   Age          714 non-null    float64
 6   SibSp        891 non-null    int64  
 7   Parch        891 non-null    int64  
 8   Ticket       891 non-null    object 
 9   Fare         891 non-null    float64
 10  Cabin        204 non-null    object 
 11  Embarked     889 non-null    object 
dtypes: float64(2), int64(5), object(5)
memory usage: 83.7+ KB
In [6]:
test.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 418 entries, 0 to 417
Data columns (total 11 columns):
 #   Column       Non-Null Count  Dtype  
---  ------       --------------  -----  
 0   PassengerId  418 non-null    int64  
 1   Pclass       418 non-null    int64  
 2   Name         418 non-null    object 
 3   Sex          418 non-null    object 
 4   Age          332 non-null    float64
 5   SibSp        418 non-null    int64  
 6   Parch        418 non-null    int64  
 7   Ticket       418 non-null    object 
 8   Fare         417 non-null    float64
 9   Cabin        91 non-null     object 
 10  Embarked     418 non-null    object 
dtypes: float64(2), int64(4), object(5)
memory usage: 36.0+ KB

02. 입력 및 출력 지정

목차로 이동하기

  • 딥러닝의 이해를 위해 일부 특징(변수)만 지정하였음.
  • 이미지를 사용할 때는 지정된 이미지 전체를 입력 데이터로 사용하는 경우가 대부분.
In [10]:
input_col = ['Pclass', 'SibSp', 'Parch']
labeled_col = ['Survived']
In [11]:
X = train[ input_col ]
y = train[ labeled_col ]
X_val = test[ input_col ]
In [12]:
seed = 0
np.random.seed(seed)
In [13]:
from sklearn.model_selection import train_test_split

X_train, X_test, y_train, y_test = train_test_split(X, y, 
                                                random_state=0)
In [14]:
print(X_train.shape, X_test.shape)
print()
print(y_train.shape, y_test.shape)

(668, 3) (223, 3)

(668, 1) (223, 1)

03. 딥러닝 구축 및 학습시키기

목차로 이동하기

In [15]:
from keras.models import Sequential
from keras.layers import Dense
In [16]:
model = Sequential()
model.add(Dense(30, input_dim=3, activation='relu'))
model.add(Dense(15, activation='relu') )
model.add(Dense(1, activation='sigmoid'))

딥러닝 설정 및 학습¶

In [17]:
model.compile(loss = 'binary_crossentropy', 
              optimizer='adam', 
              metrics=['accuracy'])
model.fit(X_train, y_train, epochs=100, batch_size=10)

Epoch 1/100
67/67 [==============================] - 1s 2ms/step - loss: 0.6674 - accuracy: 0.6048
Epoch 2/100
67/67 [==============================] - 0s 2ms/step - loss: 0.6365 - accuracy: 0.6243
Epoch 3/100
67/67 [==============================] - 0s 3ms/step - loss: 0.6274 - accuracy: 0.6213
Epoch 4/100
67/67 [==============================] - 0s 2ms/step - loss: 0.6241 - accuracy: 0.6392
Epoch 5/100
67/67 [==============================] - 0s 2ms/step - loss: 0.6185 - accuracy: 0.6302
Epoch 6/100
67/67 [==============================] - 0s 2ms/step - loss: 0.6156 - accuracy: 0.6632
Epoch 7/100
67/67 [==============================] - 0s 2ms/step - loss: 0.6150 - accuracy: 0.6437
Epoch 8/100
67/67 [==============================] - 0s 2ms/step - loss: 0.6113 - accuracy: 0.6497
Epoch 9/100
67/67 [==============================] - 0s 2ms/step - loss: 0.6097 - accuracy: 0.6602
Epoch 10/100
67/67 [==============================] - 0s 2ms/step - loss: 0.6065 - accuracy: 0.6796
Epoch 11/100
67/67 [==============================] - 0s 2ms/step - loss: 0.6062 - accuracy: 0.6841
Epoch 12/100
67/67 [==============================] - 0s 2ms/step - loss: 0.6049 - accuracy: 0.6946
Epoch 13/100
67/67 [==============================] - 0s 2ms/step - loss: 0.6046 - accuracy: 0.6856
Epoch 14/100
67/67 [==============================] - 0s 2ms/step - loss: 0.6014 - accuracy: 0.6826
Epoch 15/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5990 - accuracy: 0.6886
Epoch 16/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5982 - accuracy: 0.6901
Epoch 17/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5961 - accuracy: 0.6901
Epoch 18/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5964 - accuracy: 0.6901
Epoch 19/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5963 - accuracy: 0.6901
Epoch 20/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5959 - accuracy: 0.6916
Epoch 21/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5932 - accuracy: 0.6916
Epoch 22/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5930 - accuracy: 0.6961
Epoch 23/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5925 - accuracy: 0.6901
Epoch 24/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5925 - accuracy: 0.6961
Epoch 25/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5928 - accuracy: 0.6946
Epoch 26/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5907 - accuracy: 0.6976
Epoch 27/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5905 - accuracy: 0.6976
Epoch 28/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5893 - accuracy: 0.7021
Epoch 29/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5880 - accuracy: 0.6976
Epoch 30/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5886 - accuracy: 0.6946
Epoch 31/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5886 - accuracy: 0.6991
Epoch 32/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5876 - accuracy: 0.7021
Epoch 33/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5887 - accuracy: 0.6976
Epoch 34/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5866 - accuracy: 0.6991
Epoch 35/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5873 - accuracy: 0.6976
Epoch 36/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5857 - accuracy: 0.7021
Epoch 37/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5860 - accuracy: 0.6961
Epoch 38/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5852 - accuracy: 0.7081
Epoch 39/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5848 - accuracy: 0.6976
Epoch 40/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5849 - accuracy: 0.7036
Epoch 41/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5831 - accuracy: 0.6976
Epoch 42/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5828 - accuracy: 0.7036
Epoch 43/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5910 - accuracy: 0.7036
Epoch 44/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5843 - accuracy: 0.6991
Epoch 45/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5831 - accuracy: 0.7066
Epoch 46/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5847 - accuracy: 0.7006
Epoch 47/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5827 - accuracy: 0.7021
Epoch 48/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5846 - accuracy: 0.7036
Epoch 49/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5819 - accuracy: 0.7021
Epoch 50/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5820 - accuracy: 0.7036
Epoch 51/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5845 - accuracy: 0.7051
Epoch 52/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5820 - accuracy: 0.7096
Epoch 53/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5821 - accuracy: 0.7036
Epoch 54/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5839 - accuracy: 0.7036
Epoch 55/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5821 - accuracy: 0.7036
Epoch 56/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5812 - accuracy: 0.7081
Epoch 57/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5825 - accuracy: 0.7051
Epoch 58/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5819 - accuracy: 0.7051
Epoch 59/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5798 - accuracy: 0.7066
Epoch 60/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5813 - accuracy: 0.7081
Epoch 61/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5807 - accuracy: 0.7096
Epoch 62/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5809 - accuracy: 0.7066
Epoch 63/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5802 - accuracy: 0.7051
Epoch 64/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5820 - accuracy: 0.7066
Epoch 65/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5810 - accuracy: 0.7066
Epoch 66/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5796 - accuracy: 0.7096
Epoch 67/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5795 - accuracy: 0.7081
Epoch 68/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5806 - accuracy: 0.7081
Epoch 69/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5792 - accuracy: 0.7021
Epoch 70/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5779 - accuracy: 0.7081
Epoch 71/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5804 - accuracy: 0.7081
Epoch 72/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5784 - accuracy: 0.7096
Epoch 73/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5809 - accuracy: 0.7066
Epoch 74/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5787 - accuracy: 0.7081
Epoch 75/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5776 - accuracy: 0.7096
Epoch 76/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5752 - accuracy: 0.7081
Epoch 77/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5777 - accuracy: 0.7021
Epoch 78/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5796 - accuracy: 0.7096
Epoch 79/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5776 - accuracy: 0.7111
Epoch 80/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5795 - accuracy: 0.7066
Epoch 81/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5773 - accuracy: 0.7081
Epoch 82/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5785 - accuracy: 0.7021
Epoch 83/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5765 - accuracy: 0.7126
Epoch 84/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5766 - accuracy: 0.7111
Epoch 85/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5778 - accuracy: 0.7066
Epoch 86/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5768 - accuracy: 0.7096
Epoch 87/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5763 - accuracy: 0.7096
Epoch 88/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5767 - accuracy: 0.7111
Epoch 89/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5772 - accuracy: 0.7066
Epoch 90/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5775 - accuracy: 0.7096
Epoch 91/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5762 - accuracy: 0.7111
Epoch 92/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5767 - accuracy: 0.7051
Epoch 93/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5761 - accuracy: 0.7111
Epoch 94/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5768 - accuracy: 0.7111
Epoch 95/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5757 - accuracy: 0.7111
Epoch 96/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5774 - accuracy: 0.7066
Epoch 97/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5763 - accuracy: 0.7111
Epoch 98/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5751 - accuracy: 0.7051
Epoch 99/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5757 - accuracy: 0.7036
Epoch 100/100
67/67 [==============================] - 0s 2ms/step - loss: 0.5766 - accuracy: 0.7066
Out[17]:
<keras.callbacks.History at 0x126df06c280>

모델 평가¶

In [18]:
model.evaluate(X_test, y_test)

7/7 [==============================] - 0s 1ms/step - loss: 0.5864 - accuracy: 0.7309
Out[18]:
[0.5864036083221436, 0.7309417128562927]
In [19]:
print("\n Accuracy : %.4f" % (model.evaluate(X_test, y_test)[1]))

7/7 [==============================] - 0s 3ms/step - loss: 0.5864 - accuracy: 0.7309

 Accuracy : 0.7309
In [20]:
pred = model.predict(X_val)

14/14 [==============================] - 0s 1ms/step
In [21]:
sub = pd.read_csv("./titanic/gender_submission.csv")
sub.columns
Out[21]:
Index(['PassengerId', 'Survived'], dtype='object')
In [22]:
pred[:, 0] > 0.5
Out[22]:
array([False, False, False, False, False, False, False,  True, False,
       False, False,  True,  True, False,  True, False, False, False,
       False, False,  True,  True,  True,  True,  True, False,  True,
       False,  True, False, False, False, False, False,  True, False,
       False, False, False, False,  True,  True, False, False,  True,
       False,  True, False,  True,  True,  True, False,  True,  True,
       False, False, False, False, False,  True, False, False, False,
       False,  True, False, False,  True,  True, False, False, False,
       False,  True,  True,  True, False,  True, False, False, False,
        True,  True, False, False, False, False, False, False,  True,
       False, False,  True, False,  True, False,  True, False, False,
       False,  True, False, False, False, False, False, False, False,
       False, False, False, False,  True, False,  True, False, False,
       False,  True, False, False, False,  True, False, False,  True,
       False, False, False, False, False,  True, False, False, False,
       False, False, False, False, False, False,  True,  True, False,
        True, False,  True, False,  True,  True,  True, False, False,
        True, False, False,  True, False,  True,  True, False, False,
       False, False, False, False,  True, False,  True, False, False,
       False, False, False, False,  True, False,  True, False,  True,
       False,  True,  True, False,  True, False,  True, False, False,
       False, False,  True, False, False,  True, False,  True, False,
       False, False, False,  True,  True,  True, False,  True, False,
       False,  True, False, False, False, False, False, False,  True,
       False,  True,  True, False, False, False, False, False,  True,
        True, False, False, False, False, False,  True, False, False,
        True, False,  True, False,  True,  True,  True,  True,  True,
       False, False,  True, False,  True, False, False,  True, False,
        True, False, False, False, False, False, False, False, False,
       False,  True, False, False, False,  True, False, False, False,
        True, False,  True, False, False, False, False, False, False,
       False, False, False, False, False, False, False, False,  True,
       False, False,  True, False, False,  True, False, False,  True,
       False,  True, False, False, False,  True, False, False,  True,
        True,  True,  True, False, False, False, False, False,  True,
       False,  True, False, False, False, False, False, False,  True,
        True, False,  True,  True, False, False,  True,  True, False,
       False, False,  True, False,  True, False, False, False, False,
       False,  True, False, False, False, False, False, False,  True,
       False, False,  True, False,  True,  True, False, False, False,
       False,  True, False, False,  True, False, False, False,  True,
       False, False,  True,  True, False,  True,  True, False, False,
        True, False, False, False, False, False, False,  True, False,
       False, False, False,  True,  True,  True, False, False,  True,
       False,  True, False, False,  True, False,  True,  True,  True,
       False, False,  True, False, False, False,  True, False, False,
        True, False, False, False])
In [23]:
sub['Survived'] = pred[:, 0] > 0.5
In [24]:
sub.loc[sub['Survived']==True, 'Survived'] = 1
sub.loc[sub['Survived']==False, 'Survived'] = 0
In [26]:
sub.to_csv("titanic_submit.csv", index=False)

추가 실습¶

  • 여러개의 특징을 선택 및 신경망의 뉴런 추가 등으로 성능을 개선시켜 보자.