main.py

import warnings

import pandas as pd
from sklearn.preprocessing import MinMaxScaler
import torch.nn as nn
import torch.optim as optim

from src.features import build_features_final
from src.models import predict_model
from src.train.train import train, evaluation, prediction_to_csv
from src.data.make_dataset import DatasetLoader
from src.config.config import seed_everything, cfg

warnings.filterwarnings(action='ignore')
seed_everything(cfg.SEED)

scaler = MinMaxScaler()

drop_feature = ['type', 'motor_hp', 'air_end_temp', 'motor_rpm', 'motor_temp', 'motor_vibe', 'motor_current',
                'air_inflow', "air_flow_pressure", "current_by_vibration", "airflow_per_rotation", 'volumetric_efficiency',
                'air_to_motor_ratio']


# 데이터 전처리
train_data = pd.read_csv(r'data\raw\train_data.csv')
train_data = build_features_final.create_derived_features(train_data)
train_data = train_data.drop(drop_feature, axis=1)

test_data = pd.read_csv(r'data\raw\test_data.csv')
test_data = build_features_final.create_derived_features(test_data)
test_data = test_data.drop(drop_feature, axis=1)
test_data_raw = test_data.copy()

scaled_train_data = scaler.fit_transform(train_data)
scaled_test_data = scaler.transform(test_data)

scaled_train_data_df = pd.DataFrame(
    scaled_train_data, columns=train_data.columns)
scaled_test_data_df = pd.DataFrame(
    scaled_test_data, columns=train_data.columns)

n_features = scaled_train_data_df.shape[1]
print(n_features)

scaled_train_data = scaled_train_data_df.values
scaled_test_data = scaled_test_data_df.values


# 데이터 로더
dataloader = DatasetLoader(scaled_train_data, scaled_test_data)
train_loader, test_loader = dataloader.load

# 학습 파라미터
model = predict_model.AutoEncoder(input_dim=n_features, latent_dim=64)
criterion = nn.MSELoss()
optimizer = optim.Adam(model.parameters(), lr=1e-3)



if __name__ == "__main__":
    # 학습
    train(train_loader, model, criterion, optimizer)

    # 예측
    train_prediction, train_cosine = evaluation(train_loader, model)
    prediction, test_cosine = evaluation(test_loader, model, min(train_cosine))

    # 제출
    submission = prediction_to_csv(prediction)