딥러닝 모델 구현해 보기¶
학습 내용¶
- 정신 건강 설문 조사 데이터를 사용하여 개인의 우울증을 경험할 수 있는 요인을 탐색
- 첫번째 데이터 셋 : 자전거 공유 업체 시간대별 데이터
- 두번째 데이터 셋 : 타이타닉 데이터 셋(PyTorch)
- 세번째 데이터 셋 : 정신 건강 설문 조사 데이터
GPU 버전 PyTorch 설치¶
# 가상환경 만들기
conda create --name gpuDL python=3.10
# Jupyter Notebook
conda install -c conda-forge notebook jupyter
# 추가 설치
pip install pandas scikit-learn
# PyTorch 설치하기
# 01 Anaconda 사용 시
conda install pytorch=2.5.1 torchvision torchaudio pytorch-cuda=12.1 -c pytorch -c nvidia
# 02 pip 사용 시
pip3 install torch==2.5.1 torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
In [1]:
import torch
import sys
import numpy
import torch
print("Python version:", sys.version)
print("NumPy version:", numpy.__version__)
print("PyTorch version:", torch.__version__)
# CUDA 사용 가능 여부 확인
print(torch.cuda.is_available())
# 사용 가능한 GPU 장치 수 확인
print(torch.cuda.device_count())
Python version: 3.11.10 | packaged by Anaconda, Inc. | (main, Oct 3 2024, 07:22:26) [MSC v.1929 64 bit (AMD64)] NumPy version: 2.1.3 PyTorch version: 2.5.1+cpu False 0
| PyTorch 버전 | 권장 CUDA 버전 |
|---|---|
| 1.13.x | 11.7 |
| 2.0.x | 11.8 |
| 2.1.x -- | 12.1 |
In [2]:
import torch
import sklearn
print(torch.__version__)
print(sklearn.__version__)
2.5.1+cpu 1.5.2
In [3]:
import numpy as np
import pandas as pd
import torch
import torch.nn as nn
import torch.optim as optim
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
from sklearn.impute import SimpleImputer
In [4]:
import os
os.getcwd()
Out[4]:
'd:\\github\\DeepLearning_Basic_Class'
In [5]:
# 시드 고정
torch.manual_seed(42)
np.random.seed(42)
# 1. 데이터 준비
# 상대 경로로 데이터 로드
train = pd.read_csv('./datasets/health_mental_24/train.csv')
test = pd.read_csv('./datasets/health_mental_24/test.csv')
sub = pd.read_csv('./datasets/health_mental_24/sample_submission.csv')
# 데이터 shape 확인
print("훈련 데이터 shape:", train.shape)
print("테스트 데이터 shape:", test.shape)
# 데이터 정보 확인
print("\n훈련 데이터 정보:")
print(train.info())
print("\n테스트 데이터 정보:")
print(test.info())
훈련 데이터 shape: (140700, 20) 테스트 데이터 shape: (93800, 19) 훈련 데이터 정보: <class 'pandas.core.frame.DataFrame'> RangeIndex: 140700 entries, 0 to 140699 Data columns (total 20 columns): # Column Non-Null Count Dtype --- ------ -------------- ----- 0 id 140700 non-null int64 1 Name 140700 non-null object 2 Gender 140700 non-null object 3 Age 140700 non-null float64 4 City 140700 non-null object 5 Working Professional or Student 140700 non-null object 6 Profession 104070 non-null object 7 Academic Pressure 27897 non-null float64 8 Work Pressure 112782 non-null float64 9 CGPA 27898 non-null float64 10 Study Satisfaction 27897 non-null float64 11 Job Satisfaction 112790 non-null float64 12 Sleep Duration 140700 non-null object 13 Dietary Habits 140696 non-null object 14 Degree 140698 non-null object 15 Have you ever had suicidal thoughts ? 140700 non-null object 16 Work/Study Hours 140700 non-null float64 17 Financial Stress 140696 non-null float64 18 Family History of Mental Illness 140700 non-null object 19 Depression 140700 non-null int64 dtypes: float64(8), int64(2), object(10) memory usage: 21.5+ MB None 테스트 데이터 정보: <class 'pandas.core.frame.DataFrame'> RangeIndex: 93800 entries, 0 to 93799 Data columns (total 19 columns): # Column Non-Null Count Dtype --- ------ -------------- ----- 0 id 93800 non-null int64 1 Name 93800 non-null object 2 Gender 93800 non-null object 3 Age 93800 non-null float64 4 City 93800 non-null object 5 Working Professional or Student 93800 non-null object 6 Profession 69168 non-null object 7 Academic Pressure 18767 non-null float64 8 Work Pressure 75022 non-null float64 9 CGPA 18766 non-null float64 10 Study Satisfaction 18767 non-null float64 11 Job Satisfaction 75026 non-null float64 12 Sleep Duration 93800 non-null object 13 Dietary Habits 93795 non-null object 14 Degree 93798 non-null object 15 Have you ever had suicidal thoughts ? 93800 non-null object 16 Work/Study Hours 93800 non-null float64 17 Financial Stress 93800 non-null float64 18 Family History of Mental Illness 93800 non-null object dtypes: float64(8), int64(1), object(10) memory usage: 13.6+ MB None
In [6]:
train.head()
Out[6]:
| id | Name | Gender | Age | City | Working Professional or Student | Profession | Academic Pressure | Work Pressure | CGPA | Study Satisfaction | Job Satisfaction | Sleep Duration | Dietary Habits | Degree | Have you ever had suicidal thoughts ? | Work/Study Hours | Financial Stress | Family History of Mental Illness | Depression | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 | 0 | Aaradhya | Female | 49.0 | Ludhiana | Working Professional | Chef | NaN | 5.0 | NaN | NaN | 2.0 | More than 8 hours | Healthy | BHM | No | 1.0 | 2.0 | No | 0 |
| 1 | 1 | Vivan | Male | 26.0 | Varanasi | Working Professional | Teacher | NaN | 4.0 | NaN | NaN | 3.0 | Less than 5 hours | Unhealthy | LLB | Yes | 7.0 | 3.0 | No | 1 |
| 2 | 2 | Yuvraj | Male | 33.0 | Visakhapatnam | Student | NaN | 5.0 | NaN | 8.97 | 2.0 | NaN | 5-6 hours | Healthy | B.Pharm | Yes | 3.0 | 1.0 | No | 1 |
| 3 | 3 | Yuvraj | Male | 22.0 | Mumbai | Working Professional | Teacher | NaN | 5.0 | NaN | NaN | 1.0 | Less than 5 hours | Moderate | BBA | Yes | 10.0 | 1.0 | Yes | 1 |
| 4 | 4 | Rhea | Female | 30.0 | Kanpur | Working Professional | Business Analyst | NaN | 1.0 | NaN | NaN | 1.0 | 5-6 hours | Unhealthy | BBA | Yes | 9.0 | 4.0 | Yes | 0 |
In [7]:
test.head(3)
Out[7]:
| id | Name | Gender | Age | City | Working Professional or Student | Profession | Academic Pressure | Work Pressure | CGPA | Study Satisfaction | Job Satisfaction | Sleep Duration | Dietary Habits | Degree | Have you ever had suicidal thoughts ? | Work/Study Hours | Financial Stress | Family History of Mental Illness | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 | 140700 | Shivam | Male | 53.0 | Visakhapatnam | Working Professional | Judge | NaN | 2.0 | NaN | NaN | 5.0 | Less than 5 hours | Moderate | LLB | No | 9.0 | 3.0 | Yes |
| 1 | 140701 | Sanya | Female | 58.0 | Kolkata | Working Professional | Educational Consultant | NaN | 2.0 | NaN | NaN | 4.0 | Less than 5 hours | Moderate | B.Ed | No | 6.0 | 4.0 | No |
| 2 | 140702 | Yash | Male | 53.0 | Jaipur | Working Professional | Teacher | NaN | 4.0 | NaN | NaN | 1.0 | 7-8 hours | Moderate | B.Arch | Yes | 12.0 | 4.0 | No |
In [8]:
sub.head()
Out[8]:
| id | Depression | |
|---|---|---|
| 0 | 140700 | 0 |
| 1 | 140701 | 0 |
| 2 | 140702 | 0 |
| 3 | 140703 | 0 |
| 4 | 140704 | 0 |
In [9]:
train.columns
Out[9]:
Index(['id', 'Name', 'Gender', 'Age', 'City',
'Working Professional or Student', 'Profession', 'Academic Pressure',
'Work Pressure', 'CGPA', 'Study Satisfaction', 'Job Satisfaction',
'Sleep Duration', 'Dietary Habits', 'Degree',
'Have you ever had suicidal thoughts ?', 'Work/Study Hours',
'Financial Stress', 'Family History of Mental Illness', 'Depression'],
dtype='object')
In [40]:
import numpy as np
import pandas as pd
from sklearn.impute import SimpleImputer
from sklearn.preprocessing import LabelEncoder, StandardScaler
# 필요한 피처 선택
features = [
'Age', 'Work/Study Hours', 'Financial Stress', # 숫자형 변수
'Gender', 'Working Professional or Student', 'City',
'Sleep Duration', 'Dietary Habits', 'Degree', # 범주형 변수
'Academic Pressure', 'Work Pressure', 'CGPA',
'Study Satisfaction', 'Job Satisfaction' # 결측치 있는 숫자형 변수
]
target = 'Depression'
In [42]:
# 전처리 함수 정의
def preprocess_data(df, gubun='train'):
# 데이터프레임 복사
data = df.copy()
# 범주형 변수 인코딩
categorical_features = [
'Gender',
'Working Professional or Student',
'City',
'Sleep Duration',
'Dietary Habits',
'Degree'
]
# 라벨 인코더 초기화
le = LabelEncoder()
# 범주형 변수 인코딩
for col in categorical_features:
# NaN 값을 문자열로 변환 후 인코딩
data[col] = data[col].fillna('Unknown')
data[col] = le.fit_transform(data[col].astype(str))
# 숫자형 변수 분리
numeric_features = [
'Age', 'Work/Study Hours', 'Financial Stress',
'Academic Pressure', 'Work Pressure', 'CGPA',
'Study Satisfaction', 'Job Satisfaction'
]
# 결측값 처리
imputer = SimpleImputer(strategy='median')
# 숫자형 변수 결측값 대체
numeric_data = imputer.fit_transform(data[numeric_features])
# 스케일러로 숫자형 변수 표준화
scaler = StandardScaler()
scaled_numeric_data = scaler.fit_transform(numeric_data)
# 범주형 변수와 숫자형 변수 결합
X = np.column_stack([
scaled_numeric_data, # 표준화된 숫자형 변수
data[categorical_features].values # 인코딩된 범주형 변수
])
if gubun=="train":
# 타겟 변수 추출
y = data[target].values
return X, y
else:
return X
# 전처리 적용
X, y = preprocess_data(train)
In [43]:
# 스케일링
scaler = StandardScaler()
X = scaler.fit_transform(X)
# 데이터 분할
X_train, X_test, y_train, y_test = train_test_split(
X, y, test_size=0.2, random_state=42
)
# NumPy to PyTorch Tensor 변환
# X_train이라는 NumPy 배열을 PyTorch의 FloatTensor로 변환.
# FloatTensor는 32비트 부동 소수점 숫자로 구성된 텐서를 생성.
# 이 변환은 PyTorch 모델에서 데이터를 처리할 수 있도록 준비하는 단계
X_train = torch.FloatTensor(X_train)
X_test = torch.FloatTensor(X_test)
y_train = torch.FloatTensor(y_train).unsqueeze(1)
y_test = torch.FloatTensor(y_test).unsqueeze(1)
- 딥러닝의 이해를 위해 일부 특징(변수)만 지정하였음.
- 이미지를 사용할 때는 지정된 이미지 전체를 입력 데이터로 사용하는 경우가 대부분.
In [32]:
# 2. 신경망 모델 정의
model = nn.Sequential(
nn.Linear(14, 32),
nn.ReLU(),
nn.Dropout(0.3),
nn.Linear(32, 16),
nn.ReLU(),
nn.Dropout(0.3),
nn.Linear(16, 1),
nn.Sigmoid()
)
# 3. 모델 학습
# 손실 함수와 옵티마이저 정의
criterion = nn.BCELoss() # 이진 분류 손실 함수
optimizer = optim.Adam(model.parameters(), lr=0.001)
In [35]:
# 학습 진행
epochs = 1000
for epoch in range(epochs):
# 순전파
outputs = model(X_train)
loss = criterion(outputs, y_train)
# 정확도 계산
with torch.no_grad():
# 이진 분류의 경우
predicted = (outputs > 0.5).float()
accuracy = (predicted == y_train).float().mean()
# 역전파
optimizer.zero_grad()
loss.backward()
optimizer.step()
# 20번마다 손실과 정확도 출력
if (epoch + 1) % 20 == 0:
print(f'Epoch [{epoch+1}/{epochs}], '
f'Loss: {loss.item():.4f}, '
f'Accuracy: {accuracy.item():.4f}')
Epoch [20/1000], Loss: 0.1903, Accuracy: 0.9216 Epoch [40/1000], Loss: 0.1895, Accuracy: 0.9219 Epoch [60/1000], Loss: 0.1891, Accuracy: 0.9217 Epoch [80/1000], Loss: 0.1887, Accuracy: 0.9216 Epoch [100/1000], Loss: 0.1885, Accuracy: 0.9218 Epoch [120/1000], Loss: 0.1883, Accuracy: 0.9219 Epoch [140/1000], Loss: 0.1882, Accuracy: 0.9218 Epoch [160/1000], Loss: 0.1880, Accuracy: 0.9219 Epoch [180/1000], Loss: 0.1879, Accuracy: 0.9219 Epoch [200/1000], Loss: 0.1878, Accuracy: 0.9221 Epoch [220/1000], Loss: 0.1877, Accuracy: 0.9222 Epoch [240/1000], Loss: 0.1875, Accuracy: 0.9222 Epoch [260/1000], Loss: 0.1874, Accuracy: 0.9223 Epoch [280/1000], Loss: 0.1873, Accuracy: 0.9223 Epoch [300/1000], Loss: 0.1872, Accuracy: 0.9224 Epoch [320/1000], Loss: 0.1872, Accuracy: 0.9224 Epoch [340/1000], Loss: 0.1871, Accuracy: 0.9224 Epoch [360/1000], Loss: 0.1870, Accuracy: 0.9224 Epoch [380/1000], Loss: 0.1869, Accuracy: 0.9225 Epoch [400/1000], Loss: 0.1868, Accuracy: 0.9224 Epoch [420/1000], Loss: 0.1868, Accuracy: 0.9224 Epoch [440/1000], Loss: 0.1867, Accuracy: 0.9225 Epoch [460/1000], Loss: 0.1866, Accuracy: 0.9225 Epoch [480/1000], Loss: 0.1865, Accuracy: 0.9225 Epoch [500/1000], Loss: 0.1865, Accuracy: 0.9225 Epoch [520/1000], Loss: 0.1864, Accuracy: 0.9225 Epoch [540/1000], Loss: 0.1863, Accuracy: 0.9226 Epoch [560/1000], Loss: 0.1862, Accuracy: 0.9226 Epoch [580/1000], Loss: 0.1862, Accuracy: 0.9226 Epoch [600/1000], Loss: 0.1861, Accuracy: 0.9226 Epoch [620/1000], Loss: 0.1860, Accuracy: 0.9226 Epoch [640/1000], Loss: 0.1860, Accuracy: 0.9226 Epoch [660/1000], Loss: 0.1859, Accuracy: 0.9228 Epoch [680/1000], Loss: 0.1859, Accuracy: 0.9228 Epoch [700/1000], Loss: 0.1858, Accuracy: 0.9228 Epoch [720/1000], Loss: 0.1858, Accuracy: 0.9228 Epoch [740/1000], Loss: 0.1857, Accuracy: 0.9229 Epoch [760/1000], Loss: 0.1857, Accuracy: 0.9228 Epoch [780/1000], Loss: 0.1856, Accuracy: 0.9228 Epoch [800/1000], Loss: 0.1856, Accuracy: 0.9229 Epoch [820/1000], Loss: 0.1855, Accuracy: 0.9230 Epoch [840/1000], Loss: 0.1855, Accuracy: 0.9230 Epoch [860/1000], Loss: 0.1854, Accuracy: 0.9231 Epoch [880/1000], Loss: 0.1854, Accuracy: 0.9231 Epoch [900/1000], Loss: 0.1853, Accuracy: 0.9231 Epoch [920/1000], Loss: 0.1853, Accuracy: 0.9232 Epoch [940/1000], Loss: 0.1853, Accuracy: 0.9231 Epoch [960/1000], Loss: 0.1852, Accuracy: 0.9231 Epoch [980/1000], Loss: 0.1852, Accuracy: 0.9231 Epoch [1000/1000], Loss: 0.1851, Accuracy: 0.9231
In [37]:
# 4. 모델 평가
model.eval() # 평가 모드
with torch.no_grad():
test_outputs = model(X_test)
predicted = (test_outputs > 0.5).float()
accuracy = (predicted == y_test).float().mean()
print(f'Test Accuracy: {accuracy.item():.4f}')
Test Accuracy: 0.9231
In [38]:
# 필요한 피처 선택
features = [
'Age', 'Work/Study Hours', 'Financial Stress', # 숫자형 변수
'Gender', 'Working Professional or Student', 'City',
'Sleep Duration', 'Dietary Habits', 'Degree', # 범주형 변수
'Academic Pressure', 'Work Pressure', 'CGPA',
'Study Satisfaction', 'Job Satisfaction' # 결측치 있는 숫자형 변수
]
# 선택한 피처로 test 데이터 준비
X_predict = preprocess_data(test, gubun='test')
# 결측값 처리
X_predict = scaler.transform(X_predict)
# 스케일링
X_predict = scaler.transform(X_predict)
# PyTorch Tensor로 변환
X_predict = torch.FloatTensor(X_predict)
# 예측 수행
model.eval()
with torch.no_grad():
pred = model(X_predict)
pred = (pred > 0.5).float()
# submission 파일에 예측값 저장
sub['Depression'] = pred.numpy()
sub.to_csv('./datasets/health_mental_24/sub03.csv', index=False)
print("예측 완료 및 sub 파일 저장.")
예측 완료 및 sub 파일 저장.
In [ ]: