Second Brain

강의 3기 AI개론 7차시 — Regression

Properties1
tagsrokey, ai, notebook, basic, study

38 min read

7장 선형회귀 (Regression analysis)

  • “부록3 매트플롯립 입문”에서 한글 폰트를 올바르게 출력하기 위한 설치 방법을 설명했다. 설치 방법은 다음과 같다.
!sudo apt-get install -y fonts-nanum* | tail -n 1
!sudo fc-cache -fv
!rm -rf ~/.cache/matplotlib
debconf: unable to initialize frontend: Dialog
debconf: (No usable dialog-like program is installed, so the dialog based frontend cannot be used. at /usr/share/perl5/Debconf/FrontEnd/Dialog.pm line 78, <> line 4.)
debconf: falling back to frontend: Readline
debconf: unable to initialize frontend: Readline
debconf: (This frontend requires a controlling tty.)
debconf: falling back to frontend: Teletype
dpkg-preconfigure: unable to re-open stdin: 
Processing triggers for fontconfig (2.13.1-4.2ubuntu5) ...
/usr/share/fonts: caching, new cache contents: 0 fonts, 1 dirs
/usr/share/fonts/truetype: caching, new cache contents: 0 fonts, 3 dirs
/usr/share/fonts/truetype/humor-sans: caching, new cache contents: 1 fonts, 0 dirs
/usr/share/fonts/truetype/liberation: caching, new cache contents: 16 fonts, 0 dirs
/usr/share/fonts/truetype/nanum: caching, new cache contents: 39 fonts, 0 dirs
/usr/local/share/fonts: caching, new cache contents: 0 fonts, 0 dirs
/root/.local/share/fonts: skipping, no such directory
/root/.fonts: skipping, no such directory
/usr/share/fonts/truetype: skipping, looped directory detected
/usr/share/fonts/truetype/humor-sans: skipping, looped directory detected
/usr/share/fonts/truetype/liberation: skipping, looped directory detected
/usr/share/fonts/truetype/nanum: skipping, looped directory detected
/var/cache/fontconfig: cleaning cache directory
/root/.cache/fontconfig: not cleaning non-existent cache directory
/root/.fontconfig: not cleaning non-existent cache directory
fc-cache: succeeded

# 필요 라이브러리 설치
 
!pip install torchviz | tail -n 1
!pip install torchinfo | tail -n 1
Successfully installed nvidia-cublas-cu12-12.4.5.8 nvidia-cuda-cupti-cu12-12.4.127 nvidia-cuda-nvrtc-cu12-12.4.127 nvidia-cuda-runtime-cu12-12.4.127 nvidia-cudnn-cu12-9.1.0.70 nvidia-cufft-cu12-11.2.1.3 nvidia-curand-cu12-10.3.5.147 nvidia-cusolver-cu12-11.6.1.9 nvidia-cusparse-cu12-12.3.1.170 nvidia-nvjitlink-cu12-12.4.127 torchviz-0.0.3
Successfully installed torchinfo-1.8.0
  • 모든 설치가 끝나면 한글 폰트를 바르게 출력하기 위해 [런타임] -> **[런타임 다시시작]**을 클릭한 다음, 아래 셀부터 코드를 실행해 주십시오.
# 라이브러리 임포트
 
%matplotlib inline
import numpy as np
import matplotlib.pyplot as plt
from IPython.display import display
 
# 폰트 관련 용도
import matplotlib.font_manager as fm
 
# Colab, Linux
# 나눔 고딕 폰트의 경로 명시
path = '/usr/share/fonts/truetype/nanum/NanumGothic.ttf'
font_name = fm.FontProperties(fname=path, size=10).get_name()
 
# Window
# font_name = "NanumBarunGothic"
 
# Mac
# font_name = "AppleGothic"
# 파이토치 관련 라이브러리
import torch
from torch import nn, optim
import torch.nn.functional as F
from torchviz import make_dot
from torchinfo import summary
 
# Boston dataset
import pandas  as pd
# 기본 폰트 설정
plt.rcParams['font.family'] = font_name
 
# 기본 폰트 사이즈 변경
plt.rcParams['font.size'] = 14
 
# 기본 그래프 사이즈 변경
plt.rcParams['figure.figsize'] = (6,6)
 
# 기본 그리드 표시
# 필요에 따라 설정할 때는, plt.grid()
plt.rcParams['axes.grid'] = True
plt.rcParams["grid.linestyle"] = ":"
 
# 마이너스 기호 정상 출력
plt.rcParams['axes.unicode_minus'] = False
 
# 넘파이 부동소수점 자릿수 표시
np.set_printoptions(suppress=True, precision=4)
# warning 표시 끄기
import warnings
warnings.simplefilter('ignore')

단순 선형 회귀 분석

입력 :1 출력 :1인 선형 함수

# 난수 시드값 고정
torch.manual_seed(123)
 
# 입력 :1 출력 :1 선형 함수의 정의
l1 = nn.Linear(1, 1)
 
# 선형 함수 확인
print(l1)
# print(list(l1.parameters()))
print(list(l1.named_parameters()))
Linear(in_features=1, out_features=1, bias=True)
[('weight', Parameter containing:
tensor([[-0.4078]], requires_grad=True)), ('bias', Parameter containing:
tensor([0.0331], requires_grad=True))]

### Check named parameters
name, tensor = list(l1.named_parameters())[0]
print(name, tensor[0], tensor[0].shape)
weight tensor([-0.4078], grad_fn=<SelectBackward0>) torch.Size([1])

# 파라미터명, 파라미터 값, shape 표시
 
for param in l1.named_parameters():
    print('name: ', param[0])
    print('tensor: ', param[1])
    print('shape: ', param[1].shape)
    print("="*50)
name:  weight
tensor:  Parameter containing:
tensor([[-0.4078]], requires_grad=True)
shape:  torch.Size([1, 1])
==================================================
name:  bias
tensor:  Parameter containing:
tensor([0.0331], requires_grad=True)
shape:  torch.Size([1])
==================================================

# 초깃값 설정
nn.init.constant_(l1.weight, 2.0)
nn.init.constant_(l1.bias, 1.0)
 
# 결과 확인
print(l1.weight)
print(l1.bias)
Parameter containing:
tensor([[2.]], requires_grad=True)
Parameter containing:
tensor([1.], requires_grad=True)

# 테스트용 데이터 생성
 
# x_np를 넘파이 배열로 정의
x_np = np.arange(-2, 2.1, 1) # float64
 
# 텐서 변수화
x = torch.tensor(x_np) # float32
print("x = \n", x)
 
# (N,1) 사이즈로 변경
x = x.view(-1,1)
print("x.view(-1,1) = \n", x)
 
# 결과 확인
print(x.shape)
print(x)
x = 
 tensor([-2., -1.,  0.,  1.,  2.], dtype=torch.float64)
x.view(-1,1) = 
 tensor([[-2.],
        [-1.],
        [ 0.],
        [ 1.],
        [ 2.]], dtype=torch.float64)
torch.Size([5, 1])
tensor([[-2.],
        [-1.],
        [ 0.],
        [ 1.],
        [ 2.]], dtype=torch.float64)

입력 :2 개 출력 :1인 선형 함수

# 입력 :2, 출력:1 선형 함수 정의
l2 = nn.Linear(2, 1)
print("Initial weights and bias", "="*50)
print(l2.weight)
print(l2.bias)
print()
 
# 초깃값 설정
print("constant weights and bias", "="*50)
nn.init.constant_(l2.weight, 1.0)
nn.init.constant_(l2.bias, 2.0)
 
# 결과 확인
print(l2.weight)
print(l2.bias)
Initial weights and bias ==================================================
Parameter containing:
tensor([[-0.3512,  0.2667]], requires_grad=True)
Parameter containing:
tensor([-0.6025], requires_grad=True)

constant weights and bias ==================================================
Parameter containing:
tensor([[1., 1.]], requires_grad=True)
Parameter containing:
tensor([2.], requires_grad=True)

# 2차원 넘파이 배열
x2_np = np.array([[0, 0], [0, 1], [1, 0], [1,1]])
 
# 텐서 변수화
x2 =  torch.tensor(x2_np).float()
 
# 결과 확인
print(x2.shape)
print(x2)
 
# 함수 값 계산
y2 = l2(x2)
 
# shape 확인
print(y2.shape)
 
# 값 확인
print(y2.data)
torch.Size([4, 2])
tensor([[0., 0.],
        [0., 1.],
        [1., 0.],
        [1., 1.]])
torch.Size([4, 1])
tensor([[2.],
        [3.],
        [3.],
        [4.]])

입력 :2, 출력 :3 선형 함수 정의

# 입력 :2, 출력 :3 선형 함수 정의
 
l3 = nn.Linear(2, 3)
 
# 초깃값 설정
nn.init.constant_(l3.weight[0,:], 1.0)
nn.init.constant_(l3.weight[1,:], 2.0)
nn.init.constant_(l3.weight[2,:], 3.0)
nn.init.constant_(l3.bias, 2.0)
 
# 결과 확인
print(l3.weight)
print(l3.bias)
Parameter containing:
tensor([[1., 1.],
        [2., 2.],
        [3., 3.]], requires_grad=True)
Parameter containing:
tensor([2., 2., 2.], requires_grad=True)

# 함수 값 계산
y3 = l3(x2)
 
# shape 확인
print(y3.shape)
 
# 값 확인
print(y3.data)
torch.Size([4, 3])
tensor([[2., 2., 2.],
        [3., 4., 5.],
        [3., 4., 5.],
        [4., 6., 8.]])

클래스를 이용한 모델 정의

# 모델 정의
 
class Net(nn.Module):
    def __init__(self, n_input, n_output):
        #  부모 클래스 nn.Module 초기화
        super().__init__()
 
        # 출력층 정의
        self.l1 = nn.Linear(n_input, n_output)
 
    # 예측 함수 정의
    def forward(self, x):
        x1 = self.l1(x) # 선형 회귀
        return x1
 
# 더미 입력
inputs = torch.rand(100,1)
labels1 = torch.rand(100,1)
 
# 인스턴스 생성(１ 입력, 1 출력 선형 모델)
n_input = 1
n_output = 1
net = Net(n_input, n_output)
print(net)
print(net.l1)
print(net.l1.weight)
print(net.l1.bias)
 
Net(
  (l1): Linear(in_features=1, out_features=1, bias=True)
)
Linear(in_features=1, out_features=1, bias=True)
Parameter containing:
tensor([[0.2678]], requires_grad=True)
Parameter containing:
tensor([-0.2211], requires_grad=True)

# 예측
# torch.matmul(inputs, net.l1.weight) + net.l1.bias
outputs = net(inputs)
print("outputs = \n", outputs)
outputs = 
 tensor([[-0.1364],
        [-0.1135],
        [-0.1894],
        [ 0.0004],
        [-0.1188],
        [-0.0443],
        [ 0.0074],
        [-0.0623],
        [-0.0506],
        [ 0.0420],
        [-0.1476],
        [-0.0448],
        [-0.1468],
        [ 0.0084],
        [ 0.0197],
        [-0.2107],
        [ 0.0270],
        [-0.0233],
        [-0.0289],
        [-0.0321],
        [ 0.0241],
        [-0.1049],
        [-0.2005],
        [-0.1257],
        [-0.1815],
        [-0.0784],
        [-0.1122],
        [-0.1590],
        [-0.0994],
        [ 0.0396],
        [-0.0978],
        [-0.0830],
        [-0.1081],
        [-0.0662],
        [ 0.0321],
        [-0.0054],
        [-0.0397],
        [-0.0581],
        [-0.0557],
        [-0.0355],
        [-0.1045],
        [-0.2117],
        [-0.1700],
        [ 0.0270],
        [-0.0792],
        [-0.1957],
        [-0.0661],
        [ 0.0234],
        [-0.2138],
        [-0.1774],
        [-0.1406],
        [-0.0819],
        [-0.1185],
        [-0.1019],
        [-0.2178],
        [-0.0245],
        [ 0.0303],
        [-0.0054],
        [-0.1820],
        [-0.1952],
        [-0.0316],
        [-0.0842],
        [-0.0324],
        [-0.2181],
        [-0.0952],
        [ 0.0072],
        [-0.0251],
        [-0.0823],
        [-0.0609],
        [-0.0999],
        [-0.1608],
        [-0.1378],
        [-0.1688],
        [ 0.0240],
        [-0.0136],
        [-0.0404],
        [-0.1899],
        [ 0.0161],
        [-0.0453],
        [ 0.0054],
        [-0.1399],
        [-0.0589],
        [ 0.0435],
        [ 0.0028],
        [ 0.0201],
        [-0.1153],
        [ 0.0148],
        [-0.1921],
        [-0.0757],
        [-0.1626],
        [-0.1185],
        [-0.1215],
        [-0.0772],
        [ 0.0346],
        [-0.0210],
        [-0.0878],
        [ 0.0078],
        [-0.1558],
        [-0.0182],
        [-0.0997]], grad_fn=<AddmmBackward0>)

MSELoss 클래스를 이용한 손실 함수

criterion = nn.MSELoss()
loss = criterion(outputs, labels1)
print("loss = ", loss)
loss.backward()
loss =  tensor(0.4184, grad_fn=<MseLossBackward0>)

print(net.l1.weight.grad)
print(net.l1.bias.grad)
tensor([[-0.5785]])
tensor([-1.1541])

회귀 분석 예제: Boston dataset

# 학습용 데이터셋 준비
 
# '보스턴 데이터셋'은 현재 사이킷런 라이브러리에서 가져올 수 있지만,
# 사이킷런에서 앞으로 이 데이터를 사용할 수 없기 때문에 웹 url에서 직접 수집
 
#  Variables in order:
#  CRIM     per capita crime rate by town
#  ZN       proportion of residential land zoned for lots over 25,000 sq.ft.
#  INDUS    proportion of non-retail business acres per town
#  CHAS     Charles River dummy variable (= 1 if tract bounds river; 0 otherwise)
#  NOX      nitric oxides concentration (parts per 10 million)
#  RM       average number of rooms per dwelling
#  AGE      proportion of owner-occupied units built prior to 1940
#  DIS      weighted distances to five Boston employment centres
#  RAD      index of accessibility to radial highways
#  TAX      full-value property-tax rate per $10,000
#  PTRATIO  pupil-teacher ratio by town
#  B        1000(Bk - 0.63)^2 where Bk is the proportion of blacks by town
#  LSTAT    % lower status of the population
#  MEDV     Median value of owner-occupied homes in $1000's
 
# CRIM: 인구당 마을별 범죄율
# ZN: 25,000 평방피트를 초과하는 주거용 토지 비율
# INDUS: 마을별 비소매업 지역 비율
# CHAS: 찰스강 더미 변수 (강과 접한 지역 = 1, 그렇지 않으면 = 0)
# NOX: 질소 산화물 농도 (1000만 분의 1 단위)
# RM: 주택당 평균 방 개수
# AGE: 1940년 이전에 건축된 자가 소유 주택의 비율
# DIS: 보스턴 주요 고용 센터 5곳까지의 가중 거리
# RAD: 방사형 고속도로 접근성 지수
# TAX: $10,000당 재산세율
# PTRATIO: 마을별 학생-교사 비율
# B: 1000(Bk - 0.63)^2, 여기서 Bk는 마을별 흑인 인구 비율
# LSTAT: 저소득층 인구 비율
# MEDV: 자가 소유 주택의 중간값 ($1000 단위)
 
data_url = "http://lib.stat.cmu.edu/datasets/boston"
raw_df = pd.read_csv(data_url, sep="\s+",
                     skiprows=22, header=None)
print(raw_df.head(10))
 
          0      1      2    3      4      5     6       7    8      9     10
0    0.00632  18.00   2.31  0.0  0.538  6.575  65.2  4.0900  1.0  296.0  15.3
1  396.90000   4.98  24.00  NaN    NaN    NaN   NaN     NaN  NaN    NaN   NaN
2    0.02731   0.00   7.07  0.0  0.469  6.421  78.9  4.9671  2.0  242.0  17.8
3  396.90000   9.14  21.60  NaN    NaN    NaN   NaN     NaN  NaN    NaN   NaN
4    0.02729   0.00   7.07  0.0  0.469  7.185  61.1  4.9671  2.0  242.0  17.8
5  392.83000   4.03  34.70  NaN    NaN    NaN   NaN     NaN  NaN    NaN   NaN
6    0.03237   0.00   2.18  0.0  0.458  6.998  45.8  6.0622  3.0  222.0  18.7
7  394.63000   2.94  33.40  NaN    NaN    NaN   NaN     NaN  NaN    NaN   NaN
8    0.06905   0.00   2.18  0.0  0.458  7.147  54.2  6.0622  3.0  222.0  18.7
9  396.90000   5.33  36.20  NaN    NaN    NaN   NaN     NaN  NaN    NaN   NaN

 
x_org = np.hstack([raw_df.values[::2, :],
                   raw_df.values[1::2, :2]]) # 짝수줄 전체, 홀 수 줄 [: 2] => Features
# x_org[:10, :5]
yt = raw_df.values[1::2, 2] ## Target
feature_names = np.array(['CRIM', 'ZN', 'INDUS', 'CHAS', 'NOX',
                          'RM', 'AGE', 'DIS', 'RAD', 'TAX', 'PTRATIO','B', 'LSTAT'])
 
# 결과 확인
print('원본 데이터', x_org.shape, yt.shape)
print('항목명: ', feature_names)
원본 데이터 (506, 13) (506,)
항목명:  ['CRIM' 'ZN' 'INDUS' 'CHAS' 'NOX' 'RM' 'AGE' 'DIS' 'RAD' 'TAX' 'PTRATIO'
 'B' 'LSTAT']

x_org[:5]
feature_names == 'RM'
array([False, False, False, False, False,  True, False, False, False,
       False, False, False, False])

# 데이터 추출(RM 항목)
x = x_org[:,feature_names == 'RM']
print('추출 후', x.shape)
print(x[:5,:])
 
# 정답 데이터 y 표시
print('정답 데이터')
print(yt[:5])
추출 후 (506, 1)
[[6.575]
 [6.421]
 [7.185]
 [6.998]
 [7.147]]
정답 데이터
[24.  21.6 34.7 33.4 36.2]

# 산포도 출력
 
plt.scatter(x, yt, s=10, c='b')
plt.xlabel('Room counts')
plt.ylabel('Price')
plt.title(' Scatter plot between Room counts vs Price ')
plt.show()

png

단순 선형 회귀

## 회귀모델
 
# 입력 차원수
n_input= x.shape[1]
 
# 출력 차원수
n_output = 1
 
print(f'입력 차원수: {n_input}  출력 차원수: {n_output}')
 
# 머신러닝 모델(예측 모델)의 클래스 정의
 
class Net(nn.Module):
    def __init__(self, n_input, n_output):
        #  부모 클래스 nn.Module 초기화
        super().__init__()
 
        # 출력층 정의
        self.l1 = nn.Linear(n_input, n_output)
 
        # 초깃값을 모두 1로 설정
        # "딥러닝을 위한 수학"과 조건을 맞추기 위함
        # nn.init.constant_(self.l1.weight, 1.0)
        # nn.init.constant_(self.l1.bias, 1.0)
 
    # 예측 함수 정의
    def forward(self, x):
        x1 = self.l1(x) # 선형 회귀
        return x1
입력 차원수: 1  출력 차원수: 1

# 인스턴스 생성
# １입력 1출력 선형 모델
 
net = Net(n_input, n_output)
 
# 모델 안의 파라미터를 확인
for parameter in net.named_parameters():
    print(f'변수명: {parameter[0]}')
    print(f'변숫값: {parameter[1].data}')
print("="*50)
# 파라미터 리스트를 가져오기 위해 parameters 함수를 사용
for parameter in net.parameters():
    print(parameter)
변수명: l1.weight
변숫값: tensor([[0.4797]])
변수명: l1.bias
변숫값: tensor([-0.5425])
==================================================
Parameter containing:
tensor([[0.4797]], requires_grad=True)
Parameter containing:
tensor([-0.5425], requires_grad=True)

print(net)
Net(
  (l1): Linear(in_features=1, out_features=1, bias=True)
)

# from torchsummary import summary
from torchinfo import summary
 
summary(net, (1,), device = 'cpu')
==========================================================================================
Layer (type:depth-idx)                   Output Shape              Param #
==========================================================================================
Net                                      [1]                       --
├─Linear: 1-1                            [1]                       2
==========================================================================================
Total params: 2
Trainable params: 2
Non-trainable params: 0
Total mult-adds (Units.MEGABYTES): 0.00
==========================================================================================
Input size (MB): 0.00
Forward/backward pass size (MB): 0.00
Params size (MB): 0.00
Estimated Total Size (MB): 0.00
==========================================================================================

# 손실 함수： 평균 제곱 오차
criterion = nn.MSELoss()
 
# 학습률
lr = 0.01
 
# 최적화 함수: 경사 하강법
optimizer = optim.SGD(net.parameters(), lr=lr)
# 입력값 x와 정답 yt의 텐서 변수화
# inputs = torch.tensor(x).float()
# labels = torch.tensor(yt).float()
inputs = torch.tensor(x, dtype = torch.float32)
labels = torch.tensor(yt, dtype = torch.float32)
 
# 차원 수 확인
 
print(inputs.shape, inputs.dtype)
print(labels.shape, labels.dtype)
torch.Size([506, 1]) torch.float32
torch.Size([506]) torch.float32

# 손실 계산을 위해 labels를 (N,1) 차원의 행렬로 변환
 
labels1 = labels.view((-1, 1))
 
# 차원 수 확인
print("label shape = ", labels1.shape)
 
# 예측 계산
 
outputs = net(inputs)
print(outputs.dtype)
print(outputs.dtype)
 
 
# 손실 계산
loss = criterion(outputs, labels1)
 
# 손실 값 가져오기
print(f'{loss.item():.5f}')
label shape =  torch.Size([506, 1])
torch.float32
torch.float32
482.66223

dict(net.named_parameters())
{'l1.weight': Parameter containing:
 tensor([[0.4797]], requires_grad=True),
 'l1.bias': Parameter containing:
 tensor([-0.5425], requires_grad=True)}

# 손실을 그래프로 나타내기
from torchviz import make_dot
 
g = make_dot(loss, params=dict(net.named_parameters()))
display(g)

svg

# 예측 계산
outputs = net(inputs)
 
# 손실 계산
loss = criterion(outputs, labels1)
 
# 경사 계산
loss.backward()
 
# 경사 계산 결과를 취득 가능하도록 함
print("net.l1.weight.grad = ", net.l1.weight.grad)
print("net.l1.bias.grad = ", net.l1.bias.grad)
 
# 파라미터 수정
optimizer.step()
 
# 파라미터 확인
print("="*50)
print(net.l1.weight)
print(net.l1.bias)
 
# 경삿값 초기화
optimizer.zero_grad()
 
# 경삿값을 모두 0으로 함
print("="*50)
print(net.l1.weight.grad)
print(net.l1.bias.grad)
net.l1.weight.grad =  tensor([[-260.6448]])
net.l1.bias.grad =  tensor([-40.1214])
==================================================
Parameter containing:
tensor([[3.0861]], requires_grad=True)
Parameter containing:
tensor([-0.1413], requires_grad=True)
==================================================
None
None

경사 하강법을 이용한 학습

# 학습률
lr = 0.01
 
# 인스턴스 생성(파라미터 값 초기화)
net = Net(n_input, n_output)
 
# 손실 함수：평균 제곱 오차
criterion = nn.MSELoss()
 
# 최적화 함수 : 경사 하강법
optimizer = optim.SGD(net.parameters(), lr=lr)
 
# 반복 횟수
num_epochs = 50000
 
# 평가 결과 기록(손실 값만 기록)
history = np.zeros((0,2))
# 반복 계산 메인 루프
 
for epoch in range(num_epochs):
 
    # 경삿값 초기화
    optimizer.zero_grad()
 
    # 예측 계산
    outputs = net(inputs)
 
    # 손실 계산
    # "딥러닝을 위한 수학"에 나온 결과와 맞추기 위해 2로 나눈 값을 손실로 정의
    loss = criterion(outputs, labels1) / 2.0
 
    # 경사 계산
    loss.backward()
 
    # 파라미터 수정
    optimizer.step()
 
    # 100회 마다 도중 경과를 기록
    if ( epoch % 100 == 0):
        history = np.vstack((history, np.array([epoch, loss.item()])))
        print(f'Epoch {epoch} loss: {loss.item():.5f}')
Epoch 0 loss: 299.93127
Epoch 100 loss: 28.92968
Epoch 200 loss: 28.76028
Epoch 300 loss: 28.59489
Epoch 400 loss: 28.43344
Epoch 500 loss: 28.27583
Epoch 600 loss: 28.12196
Epoch 700 loss: 27.97174
Epoch 800 loss: 27.82510
Epoch 900 loss: 27.68194
Epoch 1000 loss: 27.54218
Epoch 1100 loss: 27.40574
Epoch 1200 loss: 27.27254
Epoch 1300 loss: 27.14251
Epoch 1400 loss: 27.01557
Epoch 1500 loss: 26.89165
Epoch 1600 loss: 26.77067
Epoch 1700 loss: 26.65256
Epoch 1800 loss: 26.53726
Epoch 1900 loss: 26.42470
Epoch 2000 loss: 26.31482
Epoch 2100 loss: 26.20755
Epoch 2200 loss: 26.10282
Epoch 2300 loss: 26.00059
Epoch 2400 loss: 25.90078
Epoch 2500 loss: 25.80334
Epoch 2600 loss: 25.70822
Epoch 2700 loss: 25.61536
Epoch 2800 loss: 25.52471
Epoch 2900 loss: 25.43621
Epoch 3000 loss: 25.34982
Epoch 3100 loss: 25.26547
Epoch 3200 loss: 25.18313
Epoch 3300 loss: 25.10275
Epoch 3400 loss: 25.02428
Epoch 3500 loss: 24.94767
Epoch 3600 loss: 24.87288
Epoch 3700 loss: 24.79987
Epoch 3800 loss: 24.72860
Epoch 3900 loss: 24.65902
Epoch 4000 loss: 24.59109
Epoch 4100 loss: 24.52477
Epoch 4200 loss: 24.46003
Epoch 4300 loss: 24.39683
Epoch 4400 loss: 24.33513
Epoch 4500 loss: 24.27490
Epoch 4600 loss: 24.21610
Epoch 4700 loss: 24.15870
Epoch 4800 loss: 24.10266
Epoch 4900 loss: 24.04795
Epoch 5000 loss: 23.99454
Epoch 5100 loss: 23.94240
Epoch 5200 loss: 23.89150
Epoch 5300 loss: 23.84181
Epoch 5400 loss: 23.79330
Epoch 5500 loss: 23.74594
Epoch 5600 loss: 23.69971
Epoch 5700 loss: 23.65458
Epoch 5800 loss: 23.61051
Epoch 5900 loss: 23.56750
Epoch 6000 loss: 23.52551
Epoch 6100 loss: 23.48451
Epoch 6200 loss: 23.44449
Epoch 6300 loss: 23.40542
Epoch 6400 loss: 23.36728
Epoch 6500 loss: 23.33005
Epoch 6600 loss: 23.29370
Epoch 6700 loss: 23.25821
Epoch 6800 loss: 23.22357
Epoch 6900 loss: 23.18975
Epoch 7000 loss: 23.15673
Epoch 7100 loss: 23.12450
Epoch 7200 loss: 23.09304
Epoch 7300 loss: 23.06232
Epoch 7400 loss: 23.03233
Epoch 7500 loss: 23.00305
Epoch 7600 loss: 22.97447
Epoch 7700 loss: 22.94658
Epoch 7800 loss: 22.91933
Epoch 7900 loss: 22.89275
Epoch 8000 loss: 22.86679
Epoch 8100 loss: 22.84145
Epoch 8200 loss: 22.81671
Epoch 8300 loss: 22.79255
Epoch 8400 loss: 22.76898
Epoch 8500 loss: 22.74596
Epoch 8600 loss: 22.72349
Epoch 8700 loss: 22.70155
Epoch 8800 loss: 22.68013
Epoch 8900 loss: 22.65923
Epoch 9000 loss: 22.63881
Epoch 9100 loss: 22.61889
Epoch 9200 loss: 22.59944
Epoch 9300 loss: 22.58045
Epoch 9400 loss: 22.56191
Epoch 9500 loss: 22.54381
Epoch 9600 loss: 22.52615
Epoch 9700 loss: 22.50890
Epoch 9800 loss: 22.49206
Epoch 9900 loss: 22.47563
Epoch 10000 loss: 22.45958
Epoch 10100 loss: 22.44391
Epoch 10200 loss: 22.42862
Epoch 10300 loss: 22.41369
Epoch 10400 loss: 22.39911
Epoch 10500 loss: 22.38488
Epoch 10600 loss: 22.37099
Epoch 10700 loss: 22.35743
Epoch 10800 loss: 22.34419
Epoch 10900 loss: 22.33126
Epoch 11000 loss: 22.31865
Epoch 11100 loss: 22.30633
Epoch 11200 loss: 22.29431
Epoch 11300 loss: 22.28257
Epoch 11400 loss: 22.27111
Epoch 11500 loss: 22.25992
Epoch 11600 loss: 22.24900
Epoch 11700 loss: 22.23834
Epoch 11800 loss: 22.22793
Epoch 11900 loss: 22.21777
Epoch 12000 loss: 22.20785
Epoch 12100 loss: 22.19816
Epoch 12200 loss: 22.18871
Epoch 12300 loss: 22.17948
Epoch 12400 loss: 22.17047
Epoch 12500 loss: 22.16167
Epoch 12600 loss: 22.15308
Epoch 12700 loss: 22.14470
Epoch 12800 loss: 22.13652
Epoch 12900 loss: 22.12853
Epoch 13000 loss: 22.12073
Epoch 13100 loss: 22.11312
Epoch 13200 loss: 22.10568
Epoch 13300 loss: 22.09842
Epoch 13400 loss: 22.09134
Epoch 13500 loss: 22.08442
Epoch 13600 loss: 22.07767
Epoch 13700 loss: 22.07108
Epoch 13800 loss: 22.06465
Epoch 13900 loss: 22.05836
Epoch 14000 loss: 22.05223
Epoch 14100 loss: 22.04624
Epoch 14200 loss: 22.04040
Epoch 14300 loss: 22.03469
Epoch 14400 loss: 22.02912
Epoch 14500 loss: 22.02369
Epoch 14600 loss: 22.01838
Epoch 14700 loss: 22.01319
Epoch 14800 loss: 22.00814
Epoch 14900 loss: 22.00320
Epoch 15000 loss: 21.99837
Epoch 15100 loss: 21.99367
Epoch 15200 loss: 21.98907
Epoch 15300 loss: 21.98459
Epoch 15400 loss: 21.98021
Epoch 15500 loss: 21.97593
Epoch 15600 loss: 21.97176
Epoch 15700 loss: 21.96769
Epoch 15800 loss: 21.96371
Epoch 15900 loss: 21.95982
Epoch 16000 loss: 21.95603
Epoch 16100 loss: 21.95233
Epoch 16200 loss: 21.94872
Epoch 16300 loss: 21.94519
Epoch 16400 loss: 21.94175
Epoch 16500 loss: 21.93839
Epoch 16600 loss: 21.93510
Epoch 16700 loss: 21.93190
Epoch 16800 loss: 21.92877
Epoch 16900 loss: 21.92572
Epoch 17000 loss: 21.92274
Epoch 17100 loss: 21.91983
Epoch 17200 loss: 21.91699
Epoch 17300 loss: 21.91422
Epoch 17400 loss: 21.91151
Epoch 17500 loss: 21.90886
Epoch 17600 loss: 21.90629
Epoch 17700 loss: 21.90377
Epoch 17800 loss: 21.90131
Epoch 17900 loss: 21.89891
Epoch 18000 loss: 21.89656
Epoch 18100 loss: 21.89428
Epoch 18200 loss: 21.89205
Epoch 18300 loss: 21.88986
Epoch 18400 loss: 21.88773
Epoch 18500 loss: 21.88565
Epoch 18600 loss: 21.88363
Epoch 18700 loss: 21.88165
Epoch 18800 loss: 21.87971
Epoch 18900 loss: 21.87782
Epoch 19000 loss: 21.87598
Epoch 19100 loss: 21.87419
Epoch 19200 loss: 21.87243
Epoch 19300 loss: 21.87071
Epoch 19400 loss: 21.86904
Epoch 19500 loss: 21.86740
Epoch 19600 loss: 21.86581
Epoch 19700 loss: 21.86425
Epoch 19800 loss: 21.86274
Epoch 19900 loss: 21.86125
Epoch 20000 loss: 21.85980
Epoch 20100 loss: 21.85839
Epoch 20200 loss: 21.85701
Epoch 20300 loss: 21.85566
Epoch 20400 loss: 21.85434
Epoch 20500 loss: 21.85306
Epoch 20600 loss: 21.85180
Epoch 20700 loss: 21.85058
Epoch 20800 loss: 21.84939
Epoch 20900 loss: 21.84822
Epoch 21000 loss: 21.84708
Epoch 21100 loss: 21.84597
Epoch 21200 loss: 21.84488
Epoch 21300 loss: 21.84382
Epoch 21400 loss: 21.84278
Epoch 21500 loss: 21.84177
Epoch 21600 loss: 21.84079
Epoch 21700 loss: 21.83983
Epoch 21800 loss: 21.83889
Epoch 21900 loss: 21.83797
Epoch 22000 loss: 21.83707
Epoch 22100 loss: 21.83620
Epoch 22200 loss: 21.83535
Epoch 22300 loss: 21.83451
Epoch 22400 loss: 21.83370
Epoch 22500 loss: 21.83291
Epoch 22600 loss: 21.83213
Epoch 22700 loss: 21.83138
Epoch 22800 loss: 21.83063
Epoch 22900 loss: 21.82991
Epoch 23000 loss: 21.82921
Epoch 23100 loss: 21.82852
Epoch 23200 loss: 21.82785
Epoch 23300 loss: 21.82720
Epoch 23400 loss: 21.82655
Epoch 23500 loss: 21.82593
Epoch 23600 loss: 21.82532
Epoch 23700 loss: 21.82473
Epoch 23800 loss: 21.82415
Epoch 23900 loss: 21.82358
Epoch 24000 loss: 21.82302
Epoch 24100 loss: 21.82249
Epoch 24200 loss: 21.82196
Epoch 24300 loss: 21.82144
Epoch 24400 loss: 21.82094
Epoch 24500 loss: 21.82045
Epoch 24600 loss: 21.81997
Epoch 24700 loss: 21.81950
Epoch 24800 loss: 21.81904
Epoch 24900 loss: 21.81860
Epoch 25000 loss: 21.81817
Epoch 25100 loss: 21.81774
Epoch 25200 loss: 21.81732
Epoch 25300 loss: 21.81692
Epoch 25400 loss: 21.81652
Epoch 25500 loss: 21.81614
Epoch 25600 loss: 21.81576
Epoch 25700 loss: 21.81540
Epoch 25800 loss: 21.81503
Epoch 25900 loss: 21.81468
Epoch 26000 loss: 21.81434
Epoch 26100 loss: 21.81401
Epoch 26200 loss: 21.81368
Epoch 26300 loss: 21.81336
Epoch 26400 loss: 21.81305
Epoch 26500 loss: 21.81275
Epoch 26600 loss: 21.81245
Epoch 26700 loss: 21.81216
Epoch 26800 loss: 21.81188
Epoch 26900 loss: 21.81161
Epoch 27000 loss: 21.81133
Epoch 27100 loss: 21.81107
Epoch 27200 loss: 21.81082
Epoch 27300 loss: 21.81056
Epoch 27400 loss: 21.81032
Epoch 27500 loss: 21.81008
Epoch 27600 loss: 21.80985
Epoch 27700 loss: 21.80962
Epoch 27800 loss: 21.80940
Epoch 27900 loss: 21.80918
Epoch 28000 loss: 21.80897
Epoch 28100 loss: 21.80877
Epoch 28200 loss: 21.80856
Epoch 28300 loss: 21.80837
Epoch 28400 loss: 21.80817
Epoch 28500 loss: 21.80799
Epoch 28600 loss: 21.80781
Epoch 28700 loss: 21.80762
Epoch 28800 loss: 21.80745
Epoch 28900 loss: 21.80728
Epoch 29000 loss: 21.80712
Epoch 29100 loss: 21.80695
Epoch 29200 loss: 21.80679
Epoch 29300 loss: 21.80664
Epoch 29400 loss: 21.80649
Epoch 29500 loss: 21.80634
Epoch 29600 loss: 21.80619
Epoch 29700 loss: 21.80605
Epoch 29800 loss: 21.80592
Epoch 29900 loss: 21.80578
Epoch 30000 loss: 21.80565
Epoch 30100 loss: 21.80552
Epoch 30200 loss: 21.80540
Epoch 30300 loss: 21.80528
Epoch 30400 loss: 21.80516
Epoch 30500 loss: 21.80504
Epoch 30600 loss: 21.80493
Epoch 30700 loss: 21.80482
Epoch 30800 loss: 21.80471
Epoch 30900 loss: 21.80461
Epoch 31000 loss: 21.80450
Epoch 31100 loss: 21.80440
Epoch 31200 loss: 21.80431
Epoch 31300 loss: 21.80421
Epoch 31400 loss: 21.80411
Epoch 31500 loss: 21.80403
Epoch 31600 loss: 21.80394
Epoch 31700 loss: 21.80385
Epoch 31800 loss: 21.80376
Epoch 31900 loss: 21.80368
Epoch 32000 loss: 21.80360
Epoch 32100 loss: 21.80352
Epoch 32200 loss: 21.80344
Epoch 32300 loss: 21.80337
Epoch 32400 loss: 21.80330
Epoch 32500 loss: 21.80322
Epoch 32600 loss: 21.80315
Epoch 32700 loss: 21.80309
Epoch 32800 loss: 21.80302
Epoch 32900 loss: 21.80295
Epoch 33000 loss: 21.80289
Epoch 33100 loss: 21.80283
Epoch 33200 loss: 21.80277
Epoch 33300 loss: 21.80271
Epoch 33400 loss: 21.80265
Epoch 33500 loss: 21.80259
Epoch 33600 loss: 21.80254
Epoch 33700 loss: 21.80249
Epoch 33800 loss: 21.80243
Epoch 33900 loss: 21.80238
Epoch 34000 loss: 21.80233
Epoch 34100 loss: 21.80228
Epoch 34200 loss: 21.80224
Epoch 34300 loss: 21.80219
Epoch 34400 loss: 21.80214
Epoch 34500 loss: 21.80210
Epoch 34600 loss: 21.80206
Epoch 34700 loss: 21.80202
Epoch 34800 loss: 21.80197
Epoch 34900 loss: 21.80193
Epoch 35000 loss: 21.80189
Epoch 35100 loss: 21.80186
Epoch 35200 loss: 21.80181
Epoch 35300 loss: 21.80178
Epoch 35400 loss: 21.80174
Epoch 35500 loss: 21.80171
Epoch 35600 loss: 21.80168
Epoch 35700 loss: 21.80164
Epoch 35800 loss: 21.80161
Epoch 35900 loss: 21.80158
Epoch 36000 loss: 21.80155
Epoch 36100 loss: 21.80152
Epoch 36200 loss: 21.80149
Epoch 36300 loss: 21.80146
Epoch 36400 loss: 21.80143
Epoch 36500 loss: 21.80140
Epoch 36600 loss: 21.80138
Epoch 36700 loss: 21.80135
Epoch 36800 loss: 21.80132
Epoch 36900 loss: 21.80130
Epoch 37000 loss: 21.80128
Epoch 37100 loss: 21.80125
Epoch 37200 loss: 21.80123
Epoch 37300 loss: 21.80120
Epoch 37400 loss: 21.80119
Epoch 37500 loss: 21.80116
Epoch 37600 loss: 21.80114
Epoch 37700 loss: 21.80112
Epoch 37800 loss: 21.80110
Epoch 37900 loss: 21.80108
Epoch 38000 loss: 21.80106
Epoch 38100 loss: 21.80105
Epoch 38200 loss: 21.80102
Epoch 38300 loss: 21.80101
Epoch 38400 loss: 21.80099
Epoch 38500 loss: 21.80097
Epoch 38600 loss: 21.80095
Epoch 38700 loss: 21.80094
Epoch 38800 loss: 21.80092
Epoch 38900 loss: 21.80091
Epoch 39000 loss: 21.80090
Epoch 39100 loss: 21.80088
Epoch 39200 loss: 21.80087
Epoch 39300 loss: 21.80085
Epoch 39400 loss: 21.80083
Epoch 39500 loss: 21.80083
Epoch 39600 loss: 21.80081
Epoch 39700 loss: 21.80080
Epoch 39800 loss: 21.80079
Epoch 39900 loss: 21.80078
Epoch 40000 loss: 21.80076
Epoch 40100 loss: 21.80075
Epoch 40200 loss: 21.80074
Epoch 40300 loss: 21.80073
Epoch 40400 loss: 21.80072
Epoch 40500 loss: 21.80071
Epoch 40600 loss: 21.80070
Epoch 40700 loss: 21.80069
Epoch 40800 loss: 21.80068
Epoch 40900 loss: 21.80067
Epoch 41000 loss: 21.80066
Epoch 41100 loss: 21.80065
Epoch 41200 loss: 21.80064
Epoch 41300 loss: 21.80063
Epoch 41400 loss: 21.80062
Epoch 41500 loss: 21.80062
Epoch 41600 loss: 21.80061
Epoch 41700 loss: 21.80060
Epoch 41800 loss: 21.80059
Epoch 41900 loss: 21.80058
Epoch 42000 loss: 21.80058
Epoch 42100 loss: 21.80057
Epoch 42200 loss: 21.80056
Epoch 42300 loss: 21.80056
Epoch 42400 loss: 21.80055
Epoch 42500 loss: 21.80054
Epoch 42600 loss: 21.80054
Epoch 42700 loss: 21.80053
Epoch 42800 loss: 21.80052
Epoch 42900 loss: 21.80052
Epoch 43000 loss: 21.80051
Epoch 43100 loss: 21.80051
Epoch 43200 loss: 21.80050
Epoch 43300 loss: 21.80050
Epoch 43400 loss: 21.80049
Epoch 43500 loss: 21.80049
Epoch 43600 loss: 21.80048
Epoch 43700 loss: 21.80047
Epoch 43800 loss: 21.80047
Epoch 43900 loss: 21.80047
Epoch 44000 loss: 21.80046
Epoch 44100 loss: 21.80046
Epoch 44200 loss: 21.80045
Epoch 44300 loss: 21.80045
Epoch 44400 loss: 21.80045
Epoch 44500 loss: 21.80044
Epoch 44600 loss: 21.80044
Epoch 44700 loss: 21.80043
Epoch 44800 loss: 21.80043
Epoch 44900 loss: 21.80043
Epoch 45000 loss: 21.80042
Epoch 45100 loss: 21.80042
Epoch 45200 loss: 21.80042
Epoch 45300 loss: 21.80041
Epoch 45400 loss: 21.80041
Epoch 45500 loss: 21.80041
Epoch 45600 loss: 21.80040
Epoch 45700 loss: 21.80040
Epoch 45800 loss: 21.80040
Epoch 45900 loss: 21.80040
Epoch 46000 loss: 21.80039
Epoch 46100 loss: 21.80039
Epoch 46200 loss: 21.80039
Epoch 46300 loss: 21.80038
Epoch 46400 loss: 21.80038
Epoch 46500 loss: 21.80038
Epoch 46600 loss: 21.80038
Epoch 46700 loss: 21.80037
Epoch 46800 loss: 21.80037
Epoch 46900 loss: 21.80037
Epoch 47000 loss: 21.80037
Epoch 47100 loss: 21.80036
Epoch 47200 loss: 21.80036
Epoch 47300 loss: 21.80036
Epoch 47400 loss: 21.80036
Epoch 47500 loss: 21.80035
Epoch 47600 loss: 21.80035
Epoch 47700 loss: 21.80035
Epoch 47800 loss: 21.80035
Epoch 47900 loss: 21.80035
Epoch 48000 loss: 21.80035
Epoch 48100 loss: 21.80035
Epoch 48200 loss: 21.80034
Epoch 48300 loss: 21.80034
Epoch 48400 loss: 21.80034
Epoch 48500 loss: 21.80034
Epoch 48600 loss: 21.80034
Epoch 48700 loss: 21.80034
Epoch 48800 loss: 21.80034
Epoch 48900 loss: 21.80033
Epoch 49000 loss: 21.80033
Epoch 49100 loss: 21.80033
Epoch 49200 loss: 21.80033
Epoch 49300 loss: 21.80033
Epoch 49400 loss: 21.80033
Epoch 49500 loss: 21.80033
Epoch 49600 loss: 21.80033
Epoch 49700 loss: 21.80032
Epoch 49800 loss: 21.80032
Epoch 49900 loss: 21.80032

# 초기 손실값과 최종 손실값
 
print(f'초기 손실값: {history[0,1]:.5f}')
print(f'최종 손실값: {history[-1,1]:.5f}')
초기 손실값: 299.93127
최종 손실값: 21.80032

# 학습 곡선 출력(손실)
# 가장 처음 요소는 제외
 
plt.plot(history[1:,0], history[1:,1], 'b')
plt.xlabel('반복 횟수')
plt.ylabel('손실')
plt.title('학습 곡선(손실)')
plt.show()

png

# 회귀 직선 산출
 
# x의 최솟값, 최댓값
xse = np.array((x.min(), x.max())).reshape(-1,1)
Xse = torch.tensor(xse).float()
 
with torch.no_grad():
  Yse = net(Xse)
 
print(Yse.numpy())
 
# 산포도와 회귀 직선 출력
 
plt.scatter(x, yt, s=10, c='b')
plt.xlabel('방 개수')
plt.ylabel('가격')
plt.plot(Xse.data, Yse.data, c='k')
plt.title('산포도와 회귀 직선')
plt.show()
[[-2.2208]
 [45.2137]]

png

중회귀 모델

# 열(LSTAT: 저소득자 비율) 추가
 
x_add = x_org[:,feature_names == 'LSTAT']
x2 = np.hstack((x, x_add))
 
# shape 표시
print(x2.shape)
 
# 입력 데이터 x 표시
print(x2[:5,:])
(506, 2)
[[6.575 4.98 ]
 [6.421 9.14 ]
 [7.185 4.03 ]
 [6.998 2.94 ]
 [7.147 5.33 ]]

# 입력 차원수=2
 
n_input = x2.shape[1]
print(n_input)
 
# 모델 인스턴스 생성
net = Net(n_input, n_output)
2

# 모델 안의 파라미터 확인
# predict.weight가 2차원으로 바뀜
 
for parameter in net.named_parameters():
    print(f'변수명: {parameter[0]}')
    print(f'변숫값: {parameter[1].data}')
변수명: l1.weight
변숫값: tensor([[-0.5142,  0.2712]])
변수명: l1.bias
변숫값: tensor([-0.2058])

# 모델의 개요 표시
 
from torchinfo import summary
summary(net, (2,))
==========================================================================================
Layer (type:depth-idx)                   Output Shape              Param #
==========================================================================================
Net                                      [1]                       --
├─Linear: 1-1                            [1]                       3
==========================================================================================
Total params: 3
Trainable params: 3
Non-trainable params: 0
Total mult-adds (Units.MEGABYTES): 0.00
==========================================================================================
Input size (MB): 0.00
Forward/backward pass size (MB): 0.00
Params size (MB): 0.00
Estimated Total Size (MB): 0.00
==========================================================================================

# 입력 변수 x2를 텐서로 변환
# labels, labels1은 이전과 같음
 
# inputs = torch.tensor(x2).float()
inputs = torch.tensor(x2, dtype = torch.float32)
 
# 초기화 처리
 
# 학습률
# lr = 0.01
lr = 0.001
 
 
# 인스턴스 생성(파라미터 값 초기화)
net = Net(n_input, n_output)
 
# 손실 함수：평균 제곱 오차
criterion = nn.MSELoss()
 
# 최적화 함수 : 경사 하강법
optimizer = optim.SGD(net.parameters(), lr=lr)
 
# 반복 횟수
num_epochs = 50000
 
# 평가 결과 기록(손실 값만 기록)
history = np.zeros((0,2))
 
# 반복 계산 메인 루프
 
for epoch in range(num_epochs):
 
    # 경삿값 초기화
    optimizer.zero_grad()
 
    # 예측 계산
    outputs = net(inputs)
 
    # 오차 계산
    # "딥러닝을 위한 수학"에 나온 결과와 맞추기 위해 2로 나눈 값을 손실로 정의
    loss = criterion(outputs, labels1) / 2.0
 
    # 경사 계산
    loss.backward()
 
    # 파라미터 수정
    optimizer.step()
 
    # 100회 마다 도중 경과를 기록
    if ( epoch % 100 == 0):
        history = np.vstack((history, np.array([epoch, loss.item()])))
        print(f'Epoch {epoch} loss: {loss.item():.5f}')
Epoch 0 loss: 529.79584
Epoch 100 loss: 29.48064
Epoch 200 loss: 16.83812
Epoch 300 loss: 15.44524
Epoch 400 loss: 15.29177
Epoch 500 loss: 15.27485
Epoch 600 loss: 15.27296
Epoch 700 loss: 15.27274
Epoch 800 loss: 15.27270
Epoch 900 loss: 15.27268
Epoch 1000 loss: 15.27265
Epoch 1100 loss: 15.27264
Epoch 1200 loss: 15.27262
Epoch 1300 loss: 15.27260
Epoch 1400 loss: 15.27258
Epoch 1500 loss: 15.27256
Epoch 1600 loss: 15.27254
Epoch 1700 loss: 15.27252
Epoch 1800 loss: 15.27250
Epoch 1900 loss: 15.27248
Epoch 2000 loss: 15.27246
Epoch 2100 loss: 15.27244
Epoch 2200 loss: 15.27242
Epoch 2300 loss: 15.27240
Epoch 2400 loss: 15.27239
Epoch 2500 loss: 15.27237
Epoch 2600 loss: 15.27235
Epoch 2700 loss: 15.27233
Epoch 2800 loss: 15.27231
Epoch 2900 loss: 15.27229
Epoch 3000 loss: 15.27227
Epoch 3100 loss: 15.27225
Epoch 3200 loss: 15.27223
Epoch 3300 loss: 15.27222
Epoch 3400 loss: 15.27220
Epoch 3500 loss: 15.27217
Epoch 3600 loss: 15.27216
Epoch 3700 loss: 15.27214
Epoch 3800 loss: 15.27212
Epoch 3900 loss: 15.27210
Epoch 4000 loss: 15.27208
Epoch 4100 loss: 15.27207
Epoch 4200 loss: 15.27205
Epoch 4300 loss: 15.27203
Epoch 4400 loss: 15.27201
Epoch 4500 loss: 15.27199
Epoch 4600 loss: 15.27197
Epoch 4700 loss: 15.27195
Epoch 4800 loss: 15.27193
Epoch 4900 loss: 15.27192
Epoch 5000 loss: 15.27189
Epoch 5100 loss: 15.27188
Epoch 5200 loss: 15.27186
Epoch 5300 loss: 15.27184
Epoch 5400 loss: 15.27182
Epoch 5500 loss: 15.27180
Epoch 5600 loss: 15.27179
Epoch 5700 loss: 15.27177
Epoch 5800 loss: 15.27175
Epoch 5900 loss: 15.27173
Epoch 6000 loss: 15.27171
Epoch 6100 loss: 15.27170
Epoch 6200 loss: 15.27168
Epoch 6300 loss: 15.27166
Epoch 6400 loss: 15.27164
Epoch 6500 loss: 15.27162
Epoch 6600 loss: 15.27160
Epoch 6700 loss: 15.27158
Epoch 6800 loss: 15.27157
Epoch 6900 loss: 15.27155
Epoch 7000 loss: 15.27153
Epoch 7100 loss: 15.27151
Epoch 7200 loss: 15.27149
Epoch 7300 loss: 15.27148
Epoch 7400 loss: 15.27146
Epoch 7500 loss: 15.27144
Epoch 7600 loss: 15.27142
Epoch 7700 loss: 15.27140
Epoch 7800 loss: 15.27139
Epoch 7900 loss: 15.27137
Epoch 8000 loss: 15.27135
Epoch 8100 loss: 15.27133
Epoch 8200 loss: 15.27131
Epoch 8300 loss: 15.27129
Epoch 8400 loss: 15.27128
Epoch 8500 loss: 15.27126
Epoch 8600 loss: 15.27124
Epoch 8700 loss: 15.27122
Epoch 8800 loss: 15.27121
Epoch 8900 loss: 15.27119
Epoch 9000 loss: 15.27117
Epoch 9100 loss: 15.27115
Epoch 9200 loss: 15.27114
Epoch 9300 loss: 15.27112
Epoch 9400 loss: 15.27110
Epoch 9500 loss: 15.27109
Epoch 9600 loss: 15.27107
Epoch 9700 loss: 15.27105
Epoch 9800 loss: 15.27103
Epoch 9900 loss: 15.27101
Epoch 10000 loss: 15.27100
Epoch 10100 loss: 15.27098
Epoch 10200 loss: 15.27096
Epoch 10300 loss: 15.27094
Epoch 10400 loss: 15.27093
Epoch 10500 loss: 15.27091
Epoch 10600 loss: 15.27089
Epoch 10700 loss: 15.27087
Epoch 10800 loss: 15.27085
Epoch 10900 loss: 15.27084
Epoch 11000 loss: 15.27082
Epoch 11100 loss: 15.27080
Epoch 11200 loss: 15.27079
Epoch 11300 loss: 15.27077
Epoch 11400 loss: 15.27075
Epoch 11500 loss: 15.27073
Epoch 11600 loss: 15.27072
Epoch 11700 loss: 15.27070
Epoch 11800 loss: 15.27068
Epoch 11900 loss: 15.27067
Epoch 12000 loss: 15.27065
Epoch 12100 loss: 15.27063
Epoch 12200 loss: 15.27061
Epoch 12300 loss: 15.27060
Epoch 12400 loss: 15.27058
Epoch 12500 loss: 15.27057
Epoch 12600 loss: 15.27055
Epoch 12700 loss: 15.27053
Epoch 12800 loss: 15.27051
Epoch 12900 loss: 15.27050
Epoch 13000 loss: 15.27048
Epoch 13100 loss: 15.27047
Epoch 13200 loss: 15.27045
Epoch 13300 loss: 15.27043
Epoch 13400 loss: 15.27041
Epoch 13500 loss: 15.27040
Epoch 13600 loss: 15.27038
Epoch 13700 loss: 15.27036
Epoch 13800 loss: 15.27035
Epoch 13900 loss: 15.27033
Epoch 14000 loss: 15.27032
Epoch 14100 loss: 15.27029
Epoch 14200 loss: 15.27028
Epoch 14300 loss: 15.27026
Epoch 14400 loss: 15.27025
Epoch 14500 loss: 15.27023
Epoch 14600 loss: 15.27021
Epoch 14700 loss: 15.27020
Epoch 14800 loss: 15.27018
Epoch 14900 loss: 15.27017
Epoch 15000 loss: 15.27015
Epoch 15100 loss: 15.27013
Epoch 15200 loss: 15.27011
Epoch 15300 loss: 15.27010
Epoch 15400 loss: 15.27008
Epoch 15500 loss: 15.27007
Epoch 15600 loss: 15.27005
Epoch 15700 loss: 15.27003
Epoch 15800 loss: 15.27002
Epoch 15900 loss: 15.27000
Epoch 16000 loss: 15.26999
Epoch 16100 loss: 15.26997
Epoch 16200 loss: 15.26995
Epoch 16300 loss: 15.26993
Epoch 16400 loss: 15.26992
Epoch 16500 loss: 15.26990
Epoch 16600 loss: 15.26989
Epoch 16700 loss: 15.26987
Epoch 16800 loss: 15.26986
Epoch 16900 loss: 15.26984
Epoch 17000 loss: 15.26982
Epoch 17100 loss: 15.26981
Epoch 17200 loss: 15.26979
Epoch 17300 loss: 15.26977
Epoch 17400 loss: 15.26976
Epoch 17500 loss: 15.26974
Epoch 17600 loss: 15.26973
Epoch 17700 loss: 15.26971
Epoch 17800 loss: 15.26969
Epoch 17900 loss: 15.26968
Epoch 18000 loss: 15.26966
Epoch 18100 loss: 15.26965
Epoch 18200 loss: 15.26963
Epoch 18300 loss: 15.26962
Epoch 18400 loss: 15.26960
Epoch 18500 loss: 15.26958
Epoch 18600 loss: 15.26957
Epoch 18700 loss: 15.26955
Epoch 18800 loss: 15.26954
Epoch 18900 loss: 15.26952
Epoch 19000 loss: 15.26951
Epoch 19100 loss: 15.26949
Epoch 19200 loss: 15.26947
Epoch 19300 loss: 15.26946
Epoch 19400 loss: 15.26944
Epoch 19500 loss: 15.26943
Epoch 19600 loss: 15.26941
Epoch 19700 loss: 15.26939
Epoch 19800 loss: 15.26938
Epoch 19900 loss: 15.26936
Epoch 20000 loss: 15.26935
Epoch 20100 loss: 15.26933
Epoch 20200 loss: 15.26932
Epoch 20300 loss: 15.26930
Epoch 20400 loss: 15.26929
Epoch 20500 loss: 15.26927
Epoch 20600 loss: 15.26926
Epoch 20700 loss: 15.26924
Epoch 20800 loss: 15.26923
Epoch 20900 loss: 15.26921
Epoch 21000 loss: 15.26919
Epoch 21100 loss: 15.26918
Epoch 21200 loss: 15.26916
Epoch 21300 loss: 15.26915
Epoch 21400 loss: 15.26913
Epoch 21500 loss: 15.26912
Epoch 21600 loss: 15.26910
Epoch 21700 loss: 15.26909
Epoch 21800 loss: 15.26907
Epoch 21900 loss: 15.26906
Epoch 22000 loss: 15.26904
Epoch 22100 loss: 15.26903
Epoch 22200 loss: 15.26901
Epoch 22300 loss: 15.26900
Epoch 22400 loss: 15.26898
Epoch 22500 loss: 15.26897
Epoch 22600 loss: 15.26895
Epoch 22700 loss: 15.26894
Epoch 22800 loss: 15.26892
Epoch 22900 loss: 15.26891
Epoch 23000 loss: 15.26889
Epoch 23100 loss: 15.26888
Epoch 23200 loss: 15.26886
Epoch 23300 loss: 15.26885
Epoch 23400 loss: 15.26883
Epoch 23500 loss: 15.26881
Epoch 23600 loss: 15.26880
Epoch 23700 loss: 15.26879
Epoch 23800 loss: 15.26877
Epoch 23900 loss: 15.26875
Epoch 24000 loss: 15.26874
Epoch 24100 loss: 15.26873
Epoch 24200 loss: 15.26871
Epoch 24300 loss: 15.26870
Epoch 24400 loss: 15.26868
Epoch 24500 loss: 15.26867
Epoch 24600 loss: 15.26865
Epoch 24700 loss: 15.26864
Epoch 24800 loss: 15.26863
Epoch 24900 loss: 15.26861
Epoch 25000 loss: 15.26859
Epoch 25100 loss: 15.26858
Epoch 25200 loss: 15.26857
Epoch 25300 loss: 15.26855
Epoch 25400 loss: 15.26854
Epoch 25500 loss: 15.26852
Epoch 25600 loss: 15.26851
Epoch 25700 loss: 15.26849
Epoch 25800 loss: 15.26848
Epoch 25900 loss: 15.26846
Epoch 26000 loss: 15.26845
Epoch 26100 loss: 15.26844
Epoch 26200 loss: 15.26842
Epoch 26300 loss: 15.26840
Epoch 26400 loss: 15.26839
Epoch 26500 loss: 15.26838
Epoch 26600 loss: 15.26836
Epoch 26700 loss: 15.26835
Epoch 26800 loss: 15.26833
Epoch 26900 loss: 15.26832
Epoch 27000 loss: 15.26830
Epoch 27100 loss: 15.26829
Epoch 27200 loss: 15.26828
Epoch 27300 loss: 15.26826
Epoch 27400 loss: 15.26825
Epoch 27500 loss: 15.26824
Epoch 27600 loss: 15.26822
Epoch 27700 loss: 15.26821
Epoch 27800 loss: 15.26819
Epoch 27900 loss: 15.26818
Epoch 28000 loss: 15.26816
Epoch 28100 loss: 15.26815
Epoch 28200 loss: 15.26814
Epoch 28300 loss: 15.26812
Epoch 28400 loss: 15.26811
Epoch 28500 loss: 15.26809
Epoch 28600 loss: 15.26808
Epoch 28700 loss: 15.26806
Epoch 28800 loss: 15.26805
Epoch 28900 loss: 15.26804
Epoch 29000 loss: 15.26802
Epoch 29100 loss: 15.26801
Epoch 29200 loss: 15.26800
Epoch 29300 loss: 15.26798
Epoch 29400 loss: 15.26797
Epoch 29500 loss: 15.26795
Epoch 29600 loss: 15.26794
Epoch 29700 loss: 15.26793
Epoch 29800 loss: 15.26791
Epoch 29900 loss: 15.26790
Epoch 30000 loss: 15.26789
Epoch 30100 loss: 15.26787
Epoch 30200 loss: 15.26786
Epoch 30300 loss: 15.26785
Epoch 30400 loss: 15.26783
Epoch 30500 loss: 15.26782
Epoch 30600 loss: 15.26780
Epoch 30700 loss: 15.26779
Epoch 30800 loss: 15.26778
Epoch 30900 loss: 15.26776
Epoch 31000 loss: 15.26775
Epoch 31100 loss: 15.26774
Epoch 31200 loss: 15.26772
Epoch 31300 loss: 15.26771
Epoch 31400 loss: 15.26769
Epoch 31500 loss: 15.26768
Epoch 31600 loss: 15.26767
Epoch 31700 loss: 15.26766
Epoch 31800 loss: 15.26764
Epoch 31900 loss: 15.26763
Epoch 32000 loss: 15.26761
Epoch 32100 loss: 15.26760
Epoch 32200 loss: 15.26759
Epoch 32300 loss: 15.26757
Epoch 32400 loss: 15.26756
Epoch 32500 loss: 15.26755
Epoch 32600 loss: 15.26753
Epoch 32700 loss: 15.26752
Epoch 32800 loss: 15.26751
Epoch 32900 loss: 15.26749
Epoch 33000 loss: 15.26748
Epoch 33100 loss: 15.26747
Epoch 33200 loss: 15.26745
Epoch 33300 loss: 15.26744
Epoch 33400 loss: 15.26743
Epoch 33500 loss: 15.26741
Epoch 33600 loss: 15.26740
Epoch 33700 loss: 15.26739
Epoch 33800 loss: 15.26738
Epoch 33900 loss: 15.26736
Epoch 34000 loss: 15.26735
Epoch 34100 loss: 15.26733
Epoch 34200 loss: 15.26732
Epoch 34300 loss: 15.26731
Epoch 34400 loss: 15.26730
Epoch 34500 loss: 15.26728
Epoch 34600 loss: 15.26727
Epoch 34700 loss: 15.26726
Epoch 34800 loss: 15.26724
Epoch 34900 loss: 15.26723
Epoch 35000 loss: 15.26722
Epoch 35100 loss: 15.26720
Epoch 35200 loss: 15.26719
Epoch 35300 loss: 15.26718
Epoch 35400 loss: 15.26717
Epoch 35500 loss: 15.26715
Epoch 35600 loss: 15.26714
Epoch 35700 loss: 15.26713
Epoch 35800 loss: 15.26712
Epoch 35900 loss: 15.26710
Epoch 36000 loss: 15.26709
Epoch 36100 loss: 15.26707
Epoch 36200 loss: 15.26707
Epoch 36300 loss: 15.26705
Epoch 36400 loss: 15.26704
Epoch 36500 loss: 15.26702
Epoch 36600 loss: 15.26701
Epoch 36700 loss: 15.26700
Epoch 36800 loss: 15.26699
Epoch 36900 loss: 15.26697
Epoch 37000 loss: 15.26696
Epoch 37100 loss: 15.26695
Epoch 37200 loss: 15.26694
Epoch 37300 loss: 15.26692
Epoch 37400 loss: 15.26691
Epoch 37500 loss: 15.26690
Epoch 37600 loss: 15.26689
Epoch 37700 loss: 15.26687
Epoch 37800 loss: 15.26686
Epoch 37900 loss: 15.26685
Epoch 38000 loss: 15.26684
Epoch 38100 loss: 15.26682
Epoch 38200 loss: 15.26681
Epoch 38300 loss: 15.26680
Epoch 38400 loss: 15.26679
Epoch 38500 loss: 15.26677
Epoch 38600 loss: 15.26676
Epoch 38700 loss: 15.26675
Epoch 38800 loss: 15.26674
Epoch 38900 loss: 15.26673
Epoch 39000 loss: 15.26671
Epoch 39100 loss: 15.26670
Epoch 39200 loss: 15.26669
Epoch 39300 loss: 15.26667
Epoch 39400 loss: 15.26666
Epoch 39500 loss: 15.26665
Epoch 39600 loss: 15.26664
Epoch 39700 loss: 15.26663
Epoch 39800 loss: 15.26661
Epoch 39900 loss: 15.26660
Epoch 40000 loss: 15.26659
Epoch 40100 loss: 15.26657
Epoch 40200 loss: 15.26656
Epoch 40300 loss: 15.26655
Epoch 40400 loss: 15.26654
Epoch 40500 loss: 15.26653
Epoch 40600 loss: 15.26651
Epoch 40700 loss: 15.26650
Epoch 40800 loss: 15.26649
Epoch 40900 loss: 15.26648
Epoch 41000 loss: 15.26647
Epoch 41100 loss: 15.26645
Epoch 41200 loss: 15.26644
Epoch 41300 loss: 15.26643
Epoch 41400 loss: 15.26642
Epoch 41500 loss: 15.26641
Epoch 41600 loss: 15.26639
Epoch 41700 loss: 15.26638
Epoch 41800 loss: 15.26637
Epoch 41900 loss: 15.26636
Epoch 42000 loss: 15.26635
Epoch 42100 loss: 15.26633
Epoch 42200 loss: 15.26632
Epoch 42300 loss: 15.26631
Epoch 42400 loss: 15.26630
Epoch 42500 loss: 15.26629
Epoch 42600 loss: 15.26627
Epoch 42700 loss: 15.26626
Epoch 42800 loss: 15.26625
Epoch 42900 loss: 15.26624
Epoch 43000 loss: 15.26623
Epoch 43100 loss: 15.26621
Epoch 43200 loss: 15.26620
Epoch 43300 loss: 15.26619
Epoch 43400 loss: 15.26618
Epoch 43500 loss: 15.26617
Epoch 43600 loss: 15.26616
Epoch 43700 loss: 15.26615
Epoch 43800 loss: 15.26613
Epoch 43900 loss: 15.26612
Epoch 44000 loss: 15.26611
Epoch 44100 loss: 15.26610
Epoch 44200 loss: 15.26609
Epoch 44300 loss: 15.26608
Epoch 44400 loss: 15.26606
Epoch 44500 loss: 15.26605
Epoch 44600 loss: 15.26604
Epoch 44700 loss: 15.26603
Epoch 44800 loss: 15.26602
Epoch 44900 loss: 15.26601
Epoch 45000 loss: 15.26600
Epoch 45100 loss: 15.26598
Epoch 45200 loss: 15.26597
Epoch 45300 loss: 15.26596
Epoch 45400 loss: 15.26595
Epoch 45500 loss: 15.26594
Epoch 45600 loss: 15.26592
Epoch 45700 loss: 15.26592
Epoch 45800 loss: 15.26590
Epoch 45900 loss: 15.26589
Epoch 46000 loss: 15.26588
Epoch 46100 loss: 15.26587
Epoch 46200 loss: 15.26586
Epoch 46300 loss: 15.26585
Epoch 46400 loss: 15.26584
Epoch 46500 loss: 15.26582
Epoch 46600 loss: 15.26581
Epoch 46700 loss: 15.26580
Epoch 46800 loss: 15.26579
Epoch 46900 loss: 15.26578
Epoch 47000 loss: 15.26577
Epoch 47100 loss: 15.26576
Epoch 47200 loss: 15.26574
Epoch 47300 loss: 15.26573
Epoch 47400 loss: 15.26572
Epoch 47500 loss: 15.26571
Epoch 47600 loss: 15.26570
Epoch 47700 loss: 15.26569
Epoch 47800 loss: 15.26568
Epoch 47900 loss: 15.26567
Epoch 48000 loss: 15.26566
Epoch 48100 loss: 15.26564
Epoch 48200 loss: 15.26563
Epoch 48300 loss: 15.26562
Epoch 48400 loss: 15.26561
Epoch 48500 loss: 15.26560
Epoch 48600 loss: 15.26559
Epoch 48700 loss: 15.26558
Epoch 48800 loss: 15.26557
Epoch 48900 loss: 15.26556
Epoch 49000 loss: 15.26554
Epoch 49100 loss: 15.26554
Epoch 49200 loss: 15.26552
Epoch 49300 loss: 15.26551
Epoch 49400 loss: 15.26550
Epoch 49500 loss: 15.26549
Epoch 49600 loss: 15.26548
Epoch 49700 loss: 15.26547
Epoch 49800 loss: 15.26546
Epoch 49900 loss: 15.26545

# 학습 곡선 출력(손실)
 
plt.plot(history[:,0], history[:,1], 'b')
plt.xlabel('반복 횟수')
plt.ylabel('손실')
plt.title('학습 곡선(손실)')
plt.show()

png

Graph View

Backlinks