train.py

import random
random.seed(1)
import numpy as np
import pandas as pd

from torch.utils.data import DataLoader
from models import *
import os
from dataset import  XWDataset
from metrics import  XWMetrics
import time
from utils import  one_hot
from torch_func import  Agent

from torch.optim import SGD, lr_scheduler, Adam


timer = time.strftime("%Y-%m-%d-%H-%M-%S", time.localtime())
SAVE_DIR="./save_{}/".format(timer)
if not os.path.exists(SAVE_DIR):
    os.makedirs(SAVE_DIR)

data_dir="./dataset"
sub=pd.read_csv("sub.csv")
EPOCH=150
BATCH_SIZE=512


DEVICE_INFO=dict(
    gpu_num=torch.cuda.device_count(),
    device_ids = range(0, torch.cuda.device_count(), 1),
    device = torch.device("cuda:0") if torch.cuda.is_available() else "cpu",
    index_cuda=0, )

NUM_CLASSES = 19

ACTIVATION="relu"


METRICS=XWMetrics()
train_data=XWDataset(os.path.join(data_dir,"sensor_train.csv"),with_label=True)
test_data=XWDataset(os.path.join(data_dir,"sensor_test.csv"),with_label=False,)
proba_t = np.zeros((len(test_data), NUM_CLASSES))
folds=5
train_data.stratifiedKFold(folds)
for fold in range(folds):
    #划分训练集和验证集 并返回验证集数据
    model=Model(num_classes=NUM_CLASSES)
    save_dir=os.path.join(SAVE_DIR,"flod_{}".format(fold))
    agent=Agent(model=model,device_info=DEVICE_INFO, save_dir=save_dir)
    earlyStopping = None

    LOSS={ "celoss":CELoss() }
    OPTIM=Adam(model.parameters(), lr=0.001, weight_decay=0.001)
    reduceLR = lr_scheduler.ReduceLROnPlateau(OPTIM, mode="max", factor=0.5, patience=8, verbose=True)
    agent.compile(loss_dict=LOSS,optimizer=OPTIM, metrics=METRICS)
    agent.summary()
    valid_X,valid_Y=train_data.get_valid_data(fold)
    valid_Y=one_hot(valid_Y,NUM_CLASSES)
    valid_data = [(valid_X[i],valid_Y[i]) for i in range(valid_X.shape[0])]

    train_generator=DataLoader(train_data,batch_size=BATCH_SIZE,shuffle=True,num_workers=0)
    agent.fit_generator(train_generator, epochs=EPOCH,
                                  validation_data=valid_data,
                            reduceLR=reduceLR,
                            earlyStopping=earlyStopping)

    agent.load_best_model()
    test_X=[test_data.data[i] for i in range(test_data.data.shape[0])]
    scores_test= agent.predict(test_X,batch_size=1024,phase="test")
    proba_t+=scores_test/5.
sub.behavior_id = np.argmax(proba_t, axis=1)
sub.to_csv(SAVE_DIR+'submit.csv', index=False)