from torch.utils.data import Dataset, DataLoader, random_split
from torch import Tensor, nn, optim
import torch
import csv
import datetime

class BaiduDataset(Dataset):

    def __init__(self, sequence_length = 6):
        self.baidu_data_origin = []
        with open('data.csv', 'r') as fp:
            csv_reader = csv.DictReader(fp)
            for row in csv_reader:
                self.baidu_data_origin.append(float(row['y']))

        # 格式化原始数据
        self.formatted_baidu_data = []
        baidu_left_data_y = 0.05 # 2018.01, 5%
        baidu_left_data_x = datetime.date(2018, 1, 1) # 2018.01
        baidu_right_data_y = 0.052 # 2023.06, 5.2%
        baidu_right_data_x = datetime.date(2023, 6, 1) # 2023.06

        k = (baidu_right_data_y - baidu_left_data_y) / (self.baidu_data_origin[-1] - self.baidu_data_origin[0])
        b = baidu_right_data_y - k * self.baidu_data_origin[-1]

        for idx, y in enumerate(self.baidu_data_origin):
            year = int(idx / 12)
            month = idx % 12
            date_value = baidu_left_data_x.replace(year=baidu_left_data_x.year+year, month=baidu_left_data_x.month+month)
            self.formatted_baidu_data.append([date_value, k * y + b])

        if self.formatted_baidu_data[-1][0] != baidu_right_data_x:
            raise ValueError('日期的值和定义的不一致')

        self.sequence_length = sequence_length

    def __len__(self):
        return len(self.formatted_baidu_data) - self.sequence_length

    def __getitem__(self, idx):
        seq_in = self.formatted_baidu_data[idx:(idx + self.sequence_length)]
        seq_out = self.formatted_baidu_data[idx + self.sequence_length]
        network_input = [[seq[1]] for seq in seq_in]
        network_output = [seq_out[1]]
        return Tensor(network_input), Tensor(network_output)

class NeuralNetwork(nn.Module):

    def __init__(self, input_item_size = 1, output_item_size = 1):
        super().__init__()
        self.input_item_size = input_item_size
        self.output_item_size = output_item_size

        # LSTM Layer:
        num_layers = 2

        self.lstm = nn.LSTM(input_item_size, output_item_size, num_layers, batch_first=True)
        self.linear = nn.Linear(output_item_size, output_item_size)

    def forward(self, input_data):
        lstm_out, _ = self.lstm(input_data)
        return self.linear(lstm_out[:, -1, :])

def show_dataset():
    batch_size = 1
    dataset_object = BaiduDataset()
    data_loader = DataLoader(dataset_object, batch_size=batch_size)
    for x, y in data_loader:
        print('x=')
        print(x[0])
        print('y=')
        print(y[0])

def train(dataloader, model, loss_fn, optimizer):
    size = len(dataloader.dataset)
    model.train()
    for batch, (X, y) in enumerate(dataloader):
        pred = model(X)
        loss = loss_fn(pred, y)

        loss.backward()
        optimizer.step()
        optimizer.zero_grad()

        if batch % 6 == 0 or (batch) * dataloader.batch_size + len(X) == size:
            loss, current = loss.item(), (batch) * dataloader.batch_size + len(X)
            print(f"loss: {loss:>7f}  [{current:>5d}/{size:>5d}]")

def test(dataloader, model, loss_fn):
    model.eval()
    size = len(dataloader.dataset)
    num_batches = len(dataloader)
    test_loss = 0
    with torch.no_grad():
        for X, y in dataloader:
            pred = model(X)
            test_loss += loss_fn(pred, y).item()
    test_loss /= num_batches
    print(f"Test Avg loss: {test_loss:>8f} \n")

def run_train():
    dataset_object = BaiduDataset()
    #train_data_size = int(len(dataset_object) * 0.90)
    #test_data_size = len(dataset_object) - train_data_size
    #train_data, test_data = random_split(dataset_object, [train_data_size, test_data_size])
    train_data = test_data = dataset_object

    batch_size = 100
    train_data_loader = DataLoader(train_data, batch_size=batch_size)
    test_data_loader = DataLoader(test_data, batch_size=batch_size)

    model = NeuralNetwork()
    loss_fn = nn.MSELoss()
    optimizer = optim.Adam(model.parameters(), lr=0.001)

    epochs = 2000
    for t in range(epochs):
        print(f"Epoch {t+1}\n-------------------------------")
        train(train_data_loader, model, loss_fn, optimizer)
        test(test_data_loader, model, loss_fn)
    print("Done!")

    # 下面的代码用来根据测试集的一个样本接着循环预测后面，不需要可以return退出
    # return
    model.eval()
    output_series = []
    with torch.no_grad():
        # 使用测试集最后一个样本输入
        for x, y in test_data_loader:
            out = model(Tensor(x[-1]).view(1, -1, len(x[-1][-1])))
            break
        for i in range(len(x[-1][1:])):
            output_series.append(x[-1][i + 1].item())
        output_series.append(out[-1][-1].item())

        for _ in range(100):
            input_data = Tensor(output_series).view(1, -1, 1)
            out = model(input_data)
            output_series.append(out[-1][-1].item())
    with open('test.csv', 'w') as fp:
        fp.write("时间序列\n")
        for o in output_series:
            fp.write(str(o) + "\n")

if __name__ == '__main__':
    run_train()