Skip to content

Commit

Permalink
데이터 분석 기법 업데이트
Browse files Browse the repository at this point in the history
  • Loading branch information
@gibum1228
gibum1228 committed Sep 2, 2022
1 parent a6598e9 commit fc6d714
Showing 1 changed file with 61 additions and 25 deletions.
86 changes: 61 additions & 25 deletions src/phenoKOR/model.py
View file
Original file line number Diff line number Diff line change
@@ -1,22 +1,22 @@
import numpy as np
import pandas as pd
import torch
import matplotlib.pyplot as plt
from prophet import Prophet
from sklearn.metrics import mean_absolute_percentage_error
from sklearn.metrics import mean_squared_error
from sklearn.metrics import r2_score
from sklearn.metrics import mean_absolute_percentage_error
from statsmodels.tsa.arima_model import ARIMA
from prophet import Prophet
import pandas as pd
import numpy as np
import os
import platform
import matplotlib.pyplot as plt
from torch.utils.data import TensorDataset # 텐서데이터셋
from torch.utils.data import DataLoader # 데이터로더
from torch.utils.data import TensorDataset # 텐서데이터셋
from statsmodels.tsa.seasonal import seasonal_decompose

import preprocessing

# 전역변수
device = torch.device("cuda" if torch.cuda.is_available() else "cpu") # 파이토치 gpu로 돌리기
ROOT, MIDDLE = preprocessing.get_info()


def fit_arima():
pass

Expand Down Expand Up @@ -136,7 +136,8 @@ def fit_LSTM():
# step, y_count: step개의 데이터로 y_count개의 데이터를 학습
option = {
"step": range(5, 11), "y_count": 1, "batch_size": [128, 256], "hidden_dim": range(5, 11),
"dropout": [0.3, 0.4, 0.5], "layer": [1, 2], "knps_name": preprocessing.get_knps_name_en(), "class_num": range(4),
"dropout": [0.3, 0.4, 0.5], "layer": [1, 2], "knps_name": preprocessing.get_knps_name_en(),
"class_num": range(4),
"epoch": 500
}

Expand Down Expand Up @@ -237,21 +238,22 @@ def R2(y, pred_y):
def MAPE(y, pred_y):
return mean_absolute_percentage_error(y, pred_y)


def model_compare():
# 각 모델별 검증지표 데이터 로드
data_arima = pd.read_csv(f"{root}data{middle}arima_final.csv", usecols=[3, 4, 5])
data_lstm = pd.read_csv(f"{root}data{middle}lstm_final.csv", usecols=[0,1,6,7,8])
data_prophet = pd.read_csv(f"{root}data{middle}prophet_final.csv", usecols=[3,4,5])
data_arima = pd.read_csv(f"{ROOT}data{MIDDLE}arima_final.csv", usecols=[3, 4, 5])
data_lstm = pd.read_csv(f"{ROOT}data{MIDDLE}lstm_final.csv", usecols=[0, 1, 6, 7, 8])
data_prophet = pd.read_csv(f"{ROOT}data{MIDDLE}prophet_final.csv", usecols=[3, 4, 5])

data_prophet['mape'] = data_prophet['mape']/100 # MAPE 스케일 맞추기
data_prophet['mape'] = data_prophet['mape'] / 100 # MAPE 스케일 맞추기

# 컬럼명 재설정
data_arima.columns = ['R2_arima', 'RMSE_arima', 'MAPE_arima']
data_prophet.columns = ['R2_prophet', 'RMSE_prophet', 'MAPE_prophet']
data_lstm.columns = ['code', 'class_num','R2_lstm', 'RMSE_lstm', 'MAPE_lstm']
data_lstm.columns = ['code', 'class_num', 'R2_lstm', 'RMSE_lstm', 'MAPE_lstm']

# 검증 지표 합치기
final = pd.concat([data_lstm,data_prophet,data_arima],axis=1)
final = pd.concat([data_lstm, data_prophet, data_arima], axis=1)

r2 = final[['R2_lstm', 'R2_prophet', 'R2_arima']]
rmse = final[['RMSE_lstm', 'RMSE_prophet', 'RMSE_arima']]
Expand All @@ -261,9 +263,11 @@ def model_compare():
max_r2 = r2.max(axis=1)
min_rmse = rmse.min(axis=1)
min_mape = mape.min(axis=1)
max_final = pd.concat([final['code'], final['class_num'], max_r2, max_r2_idx, min_rmse, min_rmse_idx, min_mape, min_mape_idx], axis=1)
max_final.columns =['code', 'class_num', 'max_r2','max_r2_idx','min_rmse','min_rmse_idx', 'min_mape','min_mape_idx']
max_final.to_csv(f"{root}data{middle}moelcompare_result.csv")
max_final = pd.concat(
[final['code'], final['class_num'], max_r2, max_r2_idx, min_rmse, min_rmse_idx, min_mape, min_mape_idx], axis=1)
max_final.columns = ['code', 'class_num', 'max_r2', 'max_r2_idx', 'min_rmse', 'min_rmse_idx', 'min_mape',
'min_mape_idx']
max_final.to_csv(f"{ROOT}data{MIDDLE}moelcompare_result.csv")

# Prophet과 Othermodel 비교하기
r2_other = final[['R2_lstm', 'R2_arima']]
Expand All @@ -275,14 +279,18 @@ def model_compare():
min_mape_other = mape_other.min(axis=1)

# Prophet 모델과 Other 모델 차이값 계산
rmse_diff = min_rmse_other-final['RMSE_prophet']
mape_diff = min_mape_other-final['MAPE_prophet']
rmse_diff = min_rmse_other - final['RMSE_prophet']
mape_diff = min_mape_other - final['MAPE_prophet']

# 최종 비교
prop_other_final = pd.concat([final['code'], final['class_num'],final['RMSE_prophet'], min_rmse_other,rmse_diff,final['MAPE_prophet'], min_mape_other,mape_diff], axis=1)
prop_other_final.columns =['code', 'class_num', 'rmse_prophet','rmse_other','rmse_diff','mape_prophet','mape_other','mape_diff']
prop_other_final = pd.concat(
[final['code'], final['class_num'], final['RMSE_prophet'], min_rmse_other, rmse_diff, final['MAPE_prophet'],
min_mape_other, mape_diff], axis=1)
prop_other_final.columns = ['code', 'class_num', 'rmse_prophet', 'rmse_other', 'rmse_diff', 'mape_prophet',
'mape_other', 'mape_diff']

prop_other_final.to_csv(f"{ROOT}data{MIDDLE}final_compare_result.csv")

prop_other_final.to_csv(f"{root}data{middle}final_compare_result.csv")

# Arima 학습 및 예측 자동화
def arima():
Expand Down Expand Up @@ -320,4 +328,32 @@ def arima():
code.append(park)

df = pd.DataFrame({'park': code, 'class': cls, 'R2': R2_list, 'RMSE': RMSE_list, 'MAPE': MAPE_list})
df.to_csv(f"{root}data{middle}arima_final.csv")
df.to_csv(f"{ROOT}data{MIDDLE}arima_final.csv")


def plot_decompose(result):
'''
시계열 분석 그래프가 너무 작게 보여서 subplot으로 크게 보이게 하는 함수
'''
fig, (ax1, ax2, ax3, ax4) = plt.subplots(4, 1, figsize=(15, 8))
result.observed.plot(legend=False, ax=ax1)
ax1.set_ylabel('Observed')
result.trend.plot(legend=False, ax=ax2)
ax2.set_ylabel('Trend')
result.seasonal.plot(legend=False, ax=ax3)
ax3.set_ylabel('Seasonal')
result.resid.plot(legend=False, ax=ax4)
ax4.set_ylabel('Residual')


def serial_compose(data_input):
'''
시계열 분해를 통해 시계열 데이터의 추세, 계절성, 주기를 확인하고자 한다.
'''
df = data_input[['date', 'avg']]
df.index = df.date
ts = df.drop("date", axis=1)

result = seasonal_decompose(ts, model='additive', period=46)

plot_decompose(result)

0 comments on commit fc6d714

Please sign in to comment.