使用paddle2.0的LSTM代码实现完荷预测

已于 2024-08-02 09:39:02 修改
阅读量424 收藏 4
点赞数 11
文章标签： lstm 人工智能 rnn
于 2024-08-02 09:38:42 首次发布
版权声明：本文为博主原创文章，遵循 CC 4.0 BY-SA 版权协议，转载请附上原文出处链接和本声明。
本文链接：https://blog.csdn.net/weixin_45037357/article/details/140863877
版权

未来数据预测

基于过去一段时间的数据序列，预测未来一天的数据（例如股票、温度等）。
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt

data=pd.read_csv("consumption.csv")
# data.head()
data.tail()
user_idrecord_dateconsumption
88548114542016/8/272441.0
88548214542016/8/282200.0
88548314542016/8/292221.0
88548414542016/8/301671.0
88548514542016/8/312176.0
data["user_id"] == 1 # 掩码
print(data["user_id"] == 1)

0          True
1          True
2          True
3          True
4          True
          ...  
885481    False
885482    False
885483    False
885484    False
885485    False
Name: user_id, Length: 885486, dtype: bool

data[data["user_id"] == 55]
user_idrecord_dateconsumption
32886552015/1/14225.0
32887552015/1/24027.0
32888552015/1/33532.0
32889552015/1/44687.0
32890552015/1/54555.0
............
33490552016/8/273664.0
33491552016/8/283169.0
33492552016/8/295182.0
33493552016/8/304720.0
33494552016/8/315116.0

609 rows × 3 columns

data.groupby('user_id')['consumption'].max()

user_id
1        4293.0
2         485.0
3        1044.0
4        7833.0
5        3100.0
         ...   
1450     1714.0
1451     3095.0
1452    11579.0
1453     2348.0
1454     3596.0
Name: consumption, Length: 1454, dtype: float64

data.groupby(["user_id"])["consumption"].min().plot()

<matplotlib.axes._subplots.AxesSubplot at 0x7f86e472ed90>

在这里插入图片描述

data.describe()
user_idconsumption
count885486.0000008.854860e+05
mean727.5000002.619980e+03
std419.7337833.154743e+04
min1.0000001.000000e+00
25%364.0000004.200000e+01
50%727.5000002.610000e+02
75%1091.0000008.250000e+02
max1454.0000001.310016e+06

使用LSTM进行预测

import paddle

# 定义LSTM模型

class LSTM(paddle.nn.Layer):
    def __init__(self, input_size=1, hidden_size=1): # input_size 输入维度，单变量预测的话这个就是1 ；hidden_size 隐藏层的维度（决定了数据输出的维度）
        super().__init__()
        self.rnn = paddle.nn.LSTM(input_size=input_size, hidden_size=hidden_size,  num_layers=2) # num_layers 网络的层数
        self.linear = paddle.nn.Linear(hidden_size, 1) # 表示output维度

    def forward(self, inputs):
        y, (hidden, cell) = self.rnn(inputs)
        output = self.linear(hidden)
        output = output[0]+output[1]
        # output = paddle.squeeze(output)  # 该OP会删除输入Tensor的Shape中尺寸为1的维度。
        return output                  

# 测试LSTM模型
input_size = 1  # 输入维度 单变量预测的话这个就是1 
seq_len = 10  # 用多少个数据预测后一个数据
batch_size = 4  # 一批4个样本长度为10

x = paddle.randn((batch_size, seq_len, input_size))
model = LSTM(input_size)
print('model:',model)
print(x.shape) # (batch_size, seq_len, input_size)
y=model.forward(x)
print(y)
print(y.shape)   

model: LSTM(
  (rnn): LSTM(1, 1, num_layers=2
    (0): RNN(
      (cell): LSTMCell(1, 1)
    )
    (1): RNN(
      (cell): LSTMCell(1, 1)
    )
  )
  (linear): Linear(in_features=1, out_features=1, dtype=float32)
)
[4, 10, 1]
Tensor(shape=[4, 1], dtype=float32, place=CPUPlace, stop_gradient=False,
       [[0.23631290],
        [0.05070865],
        [0.09258709],
        [0.11746131]])
[4, 1]

# 数据集类
class MyDataset(paddle.io.Dataset):
    """
    步骤一：继承paddle.io.Dataset类
    """
    def __init__(self, data, num_features=13, num_labels=1):
        """
        步骤二：实现构造函数，定义数据集大小

        data: numpy.Array 1维数组
        """
        super(MyDataset, self).__init__()
        self.data = data
        self.num_features = num_features
        self.num_labels = num_labels

        x = []    # x为输入
        y = []    # y为标签
        for i in range(0, len(data) - num_features - num_labels + 1):
            x.append(data[i:i+num_features])
            y.append(data[i+num_features:i+num_features+num_labels])
        print('x——1',x)
        print('y--1',y)
        self.x = np.vstack(x).reshape(-1, self.num_features, 1)   # np.vstack:按垂直方向（行顺序）堆叠数组构成一个新的数组
        self.y = np.vstack(y)
        print('x',x)
        print('y',y)
        self.x = np.array(self.x, dtype="float32")
        self.y = np.array(self.y, dtype="float32")

        self.num_samples = len(x)

    def __getitem__(self, index):
        """
        步骤三：实现__getitem__方法，定义指定index时如何获取数据，并返回单条数据（训练数据，对应的标签）
        """
        data = self.x[index]
        label = self.y[index]

        return data, label
    def __len__(self):
        """
        步骤四：实现__len__方法，返回数据集总数目
        """
        return self.num_samples

data_01 = data[data["user_id"] == 2]['consumption'].to_numpy()
data_02=MyDataset(data_01)

y [array([212.]), array([252.]), array([279.]), array([201.]), array([104.]), array([185.]), array([163.]), array([159.]), array([170.]), array([93.]), array([82.]), array([38.]), array([207.]), array([236.]), array([245.]), array([221.]), array([290.]), array([209.]), array([294.]), array([200.]), array([194.]), array([290.]), array([283.]), array([213.]), array([145.]), array([146.]), array([295.]), array([322.]), array([197.]), array([100.]), array([69.]), array([62.]), array([36.]), array([22.]), array([36.]), array([31.]), array([15.]), array([13.]), array([15.]), array([13.]), array([23.]), array([13.]), array([15.]), array([48.]), array([126.]), array([152.]), array([254.]), array([232.]), array([203.]), array([228.]), array([210.]), array([1.]), array([236.]), array([205.]), array([275.]), array([326.]), array([244.]), array([179.]), array([114.]), array([161.]), array([184.]), array([199.]), array([209.]), array([242.]), array([272.]), array([265.]), array([220.]), array([255.]), array([189.]), array([188.]), array([137.]), array([190.]), array([164.]), array([106.]), array([30.]), array([163.]), array([159.]), array([232.]), array([161.]), array([157.]), array([200.]), array([29.]), array([24.]), array([217.]), array([176.]), array([196.]), array([140.]), array([129.]), array([25.]), array([136.]), array([88.]), array([110.]), array([262.]), array([232.]), array([152.]), array([44.]), array([42.]), array([180.]), array([153.]), array([167.]), array([286.]), array([319.]), array([131.]), array([241.]), array([70.]), array([147.]), array([267.]), array([25.]), array([53.]), array([223.]), array([144.]), array([301.]), array([208.]), array([275.]), array([296.]), array([366.]), array([252.]), array([334.]), array([321.]), array([341.]), array([450.]), array([315.]), array([349.]), array([119.]), array([425.]), array([315.]), array([448.]), array([307.]), array([294.]), array([327.]), array([300.]), array([276.]), array([276.]), array([346.]), array([388.]), array([399.]), array([397.]), array([297.]), array([320.]), array([297.]), array([189.]), array([192.]), array([178.]), array([354.]), array([318.]), array([317.]), array([286.]), array([305.]), array([410.]), array([482.]), array([440.]), array([273.]), array([184.]), array([249.]), array([276.]), array([311.]), array([294.]), array([158.]), array([263.]), array([265.]), array([200.]), array([180.]), array([343.]), array([218.]), array([258.]), array([180.]), array([190.]), array([207.]), array([238.]), array([276.]), array([240.]), array([321.]), array([254.]), array([255.]), array([296.]), array([304.]), array([208.]), array([249.]), array([339.]), array([239.]), array([326.]), array([315.]), array([345.]), array([313.]), array([312.]), array([264.]), array([234.]), array([311.]), array([332.]), array([315.]), array([321.]), array([365.]), array([304.]), array([326.]), array([412.]), array([370.]), array([365.]), array([407.]), array([386.]), array([387.]), array([316.]), array([345.]), array([359.]), array([351.]), array([428.]), array([423.]), array([335.]), array([310.]), array([316.]), array([311.]), array([294.]), array([295.]), array([331.]), array([306.]), array([261.]), array([258.]), array([257.]), array([302.]), array([287.]), array([294.]), array([256.]), array([216.]), array([250.]), array([263.]), array([304.]), array([329.]), array([48.]), array([300.]), array([330.]), array([373.]), array([290.]), array([240.]), array([55.]), array([265.]), array([233.]), array([190.]), array([152.]), array([239.]), array([241.]), array([239.]), array([214.]), array([130.]), array([112.]), array([173.]), array([155.]), array([188.]), array([158.]), array([190.]), array([235.]), array([122.]), array([190.]), array([132.]), array([147.]), array([128.]), array([112.]), array([99.]), array([24.]), array([136.]), array([243.]), array([132.]), array([14.]), array([15.]), array([202.]), array([217.]), array([109.]), array([233.]), array([244.]), array([150.]), array([243.]), array([167.]), array([258.]), array([140.]), array([170.]), array([166.]), array([119.]), array([145.]), array([186.]), array([134.]), array([142.]), array([189.]), array([196.]), array([231.]), array([211.]), array([222.]), array([153.]), array([234.]), array([320.]), array([317.]), array([322.]), array([326.]), array([265.]), array([181.]), array([273.]), array([233.]), array([264.]), array([140.]), array([209.]), array([283.]), array([180.]), array([187.]), array([195.]), array([267.]), array([324.]), array([168.]), array([165.]), array([67.]), array([112.]), array([147.]), array([57.]), array([103.]), array([130.]), array([158.]), array([154.]), array([173.]), array([159.]), array([233.]), array([175.]), array([181.]), array([188.]), array([240.]), array([192.]), array([91.]), array([102.]), array([420.]), array([452.]), array([321.]), array([250.]), array([272.]), array([258.]), array([174.]), array([142.]), array([198.]), array([191.]), array([230.]), array([258.]), array([239.]), array([214.]), array([210.]), array([189.]), array([122.]), array([46.]), array([92.]), array([76.]), array([92.]), array([251.]), array([167.]), array([249.]), array([46.]), array([121.]), array([212.]), array([118.]), array([174.]), array([34.]), array([47.]), array([148.]), array([196.]), array([294.]), array([290.]), array([352.]), array([383.]), array([305.]), array([262.]), array([265.]), array([373.]), array([285.]), array([295.]), array([365.]), array([351.]), array([320.]), array([305.]), array([271.]), array([313.]), array([348.]), array([327.]), array([243.]), array([96.]), array([334.]), array([357.]), array([283.]), array([230.]), array([297.]), array([294.]), array([213.]), array([110.]), array([169.]), array([54.]), array([89.]), array([55.]), array([62.]), array([30.]), array([26.]), array([12.]), array([12.]), array([12.]), array([12.]), array([11.]), array([15.]), array([64.]), array([177.]), array([179.]), array([217.]), array([257.]), array([354.]), array([353.]), array([415.]), array([464.]), array([381.]), array([428.]), array([455.]), array([372.]), array([362.]), array([385.]), array([383.]), array([417.]), array([285.]), array([332.]), array([339.]), array([217.]), array([254.]), array([283.]), array([307.]), array([399.]), array([320.]), array([212.]), array([164.]), array([174.]), array([313.]), array([294.]), array([284.]), array([321.]), array([247.]), array([174.]), array([287.]), array([346.]), array([377.]), array([328.]), array([331.]), array([327.]), array([276.]), array([261.]), array([268.]), array([290.]), array([256.]), array([273.]), array([293.]), array([220.]), array([192.]), array([46.]), array([301.]), array([324.]), array([288.]), array([264.]), array([185.]), array([266.]), array([278.]), array([307.]), array([278.]), array([258.]), array([313.]), array([245.]), array([342.]), array([268.]), array([329.]), array([323.]), array([247.]), array([246.]), array([210.]), array([278.]), array([254.]), array([304.]), array([40.]), array([245.]), array([246.]), array([40.]), array([277.]), array([389.]), array([235.]), array([302.]), array([313.]), array([295.]), array([148.]), array([232.]), array([244.]), array([229.]), array([268.]), array([250.]), array([212.]), array([212.]), array([296.]), array([278.]), array([214.]), array([241.]), array([288.]), array([356.]), array([342.]), array([372.]), array([337.]), array([354.]), array([317.]), array([293.]), array([327.]), array([172.]), array([155.]), array([302.]), array([101.]), array([302.]), array([298.]), array([337.]), array([231.]), array([282.]), array([330.]), array([302.]), array([339.]), array([246.]), array([339.]), array([273.]), array([271.]), array([360.]), array([321.]), array([363.]), array([351.]), array([362.]), array([312.]), array([319.]), array([307.]), array([315.]), array([328.]), array([277.]), array([200.]), array([86.]), array([233.]), array([241.]), array([253.]), array([294.]), array([285.]), array([297.]), array([173.]), array([191.]), array([324.]), array([209.]), array([192.]), array([357.]), array([331.]), array([299.]), array([373.]), array([403.]), array([344.]), array([322.]), array([338.]), array([356.]), array([208.]), array([319.]), array([346.]), array([322.]), array([370.]), array([403.]), array([359.]), array([304.]), array([277.]), array([257.]), array([263.]), array([284.]), array([275.]), array([329.]), array([249.]), array([310.]), array([285.]), array([377.]), array([396.]), array([331.]), array([361.]), array([340.]), array([375.]), array([353.]), array([340.]), array([298.]), array([328.]), array([296.]), array([324.]), array([398.]), array([400.]), array([366.]), array([485.]), array([338.]), array([297.]), array([276.]), array([385.]), array([373.]), array([309.]), array([333.]), array([261.]), array([348.]), array([294.]), array([324.]), array([343.]), array([353.])]

# 从文件读取数据集
data = pd.read_csv("consumption.csv")
data.head()
data = data[data["user_id"] == 2]['consumption'].to_numpy()

data_min = data.min()
data_max = data.max()
data = (data - data_min) / (data_max - data_min) # 归一化处理（0,1）

n_train_samples = int(len(data) * 0.85)

train_data = data[:n_train_samples]
test_data = data[n_train_samples:]
plt.plot(range(n_train_samples), train_data)
plt.plot(range(n_train_samples, len(data)), test_data)

[<matplotlib.lines.Line2D at 0x7f86e465e610>]

在这里插入图片描述

# 经过测验，num_features = 12 batch_size = 8 时效果较好
train_dataset = MyDataset(train_data, num_features=12)
test_dataset = MyDataset(test_data, num_features=12)
train_loader = paddle.io.DataLoader(train_dataset, batch_size=8, shuffle=False, drop_last=True)

y [array([0.55785124]), array([0.74173554]), array([0.66115702]), array([0.74793388]), array([0.7231405]), array([0.74586777]), array([0.64256198]), array([0.65702479]), array([0.6322314]), array([0.64876033]), array([0.67561983]), array([0.57024793]), array([0.41115702]), array([0.17561983]), array([0.47933884]), array([0.49586777]), array([0.52066116]), array([0.6053719]), array([0.58677686]), array([0.61157025]), array([0.3553719]), array([0.39256198]), array([0.66735537]), array([0.42975207]), array([0.3946281]), array([0.73553719]), array([0.68181818]), array([0.61570248]), array([0.76859504]), array([0.83057851]), array([0.70867769]), array([0.66322314]), array([0.69628099]), array([0.73347107]), array([0.42768595]), array([0.65702479]), array([0.71280992]), array([0.66322314]), array([0.76239669]), array([0.83057851]), array([0.73966942]), array([0.62603306]), array([0.57024793]), array([0.52892562]), array([0.54132231]), array([0.58471074]), array([0.5661157]), array([0.67768595]), array([0.51239669]), array([0.63842975]), array([0.58677686]), array([0.7768595]), array([0.8161157]), array([0.68181818]), array([0.74380165]), array([0.70041322]), array([0.77272727]), array([0.72727273]), array([0.70041322]), array([0.61363636]), array([0.67561983]), array([0.60950413]), array([0.66735537]), array([0.82024793]), array([0.82438017]), array([0.75413223]), array([1.]), array([0.69628099]), array([0.61157025]), array([0.56818182]), array([0.79338843]), array([0.76859504]), array([0.63636364]), array([0.68595041]), array([0.53719008]), array([0.71694215]), array([0.6053719]), array([0.66735537]), array([0.70661157]), array([0.72727273])]

paddle提供了两种训练方式

  • 一种是基础API常规的训练方式
  • 另一种是用paddle.Model对模型进行封装，通过高层API如Model.fit()、Model.evaluate()、Model.predict()等完成模型的训练与预测
# 训练方式一：
model = LSTM()
loss_fn = paddle.nn.MSELoss(reduction='mean') # reduction(缩小)，使用均值的方式
optimizer = paddle.optimizer.Adam(learning_rate=0.01,parameters=model.parameters())

for epoch in range(30):
    for batch, (batch_x, batch_y) in enumerate(train_loader()):
        # print(batch_x.shape)
        print("batch_y的shape",batch_y.shape)
        y_pred = model(batch_x)
        print("y_pred的shape",y_pred.shape)
        
        loss = loss_fn(y_pred, batch_y)
        loss.backward()
        optimizer.step()
        optimizer.clear_grad()
    print("epoch {} loss {:.4f}".format(epoch, float(loss)))

# 使用验证集进行验证
test_x = paddle.to_tensor(test_dataset.x, dtype="float32")
y_pred = model(test_x)
plt.plot(test_dataset.y)
plt.plot(y_pred.numpy())
plt.show()

epoch 29 loss 0.0188

在这里插入图片描述

# 训练方式二：
model = paddle.Model(LSTM())
model.prepare(optimizer=paddle.optimizer.Adam(learning_rate=0.01, parameters=model.parameters()),
              loss=paddle.nn.MSELoss(reduction='mean'))
model.fit(train_dataset,
          epochs=30,
          batch_size=12,
          shuffle=False,  # True 和 False 对结果影响很大
          drop_last=True)
          
# 使用验证集进行验证
y_pred = model.predict(test_dataset)
print('test_dataset.x.shape',test_dataset.x.shape)
print('test_dataset.y.shape',test_dataset.y.shape)
plt.plot(test_dataset.y)
plt.plot(np.concatenate(np.concatenate(y_pred)))
shape)
print('test_dataset.y.shape',test_dataset.y.shape)
plt.plot(test_dataset.y)
plt.plot(np.concatenate(np.concatenate(y_pred)))
plt.show()

The loss value printed in the log is the current step, and the metric is the average value of previous step.
Epoch 1/30
step 10/42 - loss: 0.0985 - 4ms/step
step 20/42 - loss: 0.0290 - 3ms/step
step 30/42 - loss: 0.0988 - 3ms/step
step 40/42 - loss: 0.0429 - 3ms/step
step 42/42 - loss: 0.0506 - 3ms/step
Epoch 2/30
step 10/42 - loss: 0.0566 - 3ms/step
step 20/42 - loss: 0.0285 - 3ms/step
step 30/42 - loss: 0.0720 - 3ms/step
step 40/42 - loss: 0.0431 - 3ms/step
step 42/42 - loss: 0.0422 - 3ms/step
Epoch 3/30
step 10/42 - loss: 0.0508 - 3ms/step
step 20/42 - loss: 0.0278 - 3ms/step
step 30/42 - loss: 0.0629 - 3ms/step
step 40/42 - loss: 0.0443 - 3ms/step
step 42/42 - loss: 0.0366 - 3ms/step
Epoch 4/30
step 10/42 - loss: 0.0464 - 3ms/step
step 20/42 - loss: 0.0279 - 3ms/step
step 30/42 - loss: 0.0529 - 3ms/step
step 40/42 - loss: 0.0470 - 3ms/step
step 42/42 - loss: 0.0296 - 3ms/step
Epoch 5/30
step 10/42 - loss: 0.0418 - 3ms/step
step 20/42 - loss: 0.0293 - 3ms/step
step 30/42 - loss: 0.0424 - 3ms/step
step 40/42 - loss: 0.0515 - 3ms/step
step 42/42 - loss: 0.0217 - 3ms/step
Epoch 6/30
step 10/42 - loss: 0.0383 - 3ms/step
step 20/42 - loss: 0.0321 - 3ms/step
step 30/42 - loss: 0.0360 - 3ms/step
step 40/42 - loss: 0.0561 - 3ms/step
step 42/42 - loss: 0.0159 - 3ms/step
Epoch 7/30
step 10/42 - loss: 0.0371 - 3ms/step
step 20/42 - loss: 0.0348 - 3ms/step
step 30/42 - loss: 0.0350 - 3ms/step
step 40/42 - loss: 0.0588 - 3ms/step
step 42/42 - loss: 0.0139 - 3ms/step
Epoch 8/30
step 10/42 - loss: 0.0374 - 3ms/step
step 20/42 - loss: 0.0362 - 3ms/step
step 30/42 - loss: 0.0354 - 3ms/step
step 40/42 - loss: 0.0599 - 3ms/step
step 42/42 - loss: 0.0136 - 3ms/step
Epoch 9/30
step 10/42 - loss: 0.0378 - 3ms/step
step 20/42 - loss: 0.0366 - 3ms/step
step 30/42 - loss: 0.0355 - 3ms/step
step 40/42 - loss: 0.0602 - 3ms/step
step 42/42 - loss: 0.0136 - 3ms/step
Epoch 10/30
step 10/42 - loss: 0.0380 - 3ms/step
step 20/42 - loss: 0.0367 - 3ms/step
step 30/42 - loss: 0.0354 - 3ms/step
step 40/42 - loss: 0.0603 - 3ms/step
step 42/42 - loss: 0.0136 - 3ms/step
Epoch 11/30
step 10/42 - loss: 0.0381 - 3ms/step
step 20/42 - loss: 0.0366 - 3ms/step
step 30/42 - loss: 0.0354 - 3ms/step
step 40/42 - loss: 0.0603 - 3ms/step
step 42/42 - loss: 0.0136 - 3ms/step
Epoch 12/30
step 10/42 - loss: 0.0382 - 3ms/step
step 20/42 - loss: 0.0366 - 3ms/step
step 30/42 - loss: 0.0353 - 3ms/step
step 40/42 - loss: 0.0603 - 3ms/step
step 42/42 - loss: 0.0136 - 3ms/step
Epoch 13/30
step 10/42 - loss: 0.0383 - 3ms/step
step 20/42 - loss: 0.0366 - 3ms/step
step 30/42 - loss: 0.0353 - 3ms/step
step 40/42 - loss: 0.0603 - 3ms/step
step 42/42 - loss: 0.0136 - 3ms/step
Epoch 14/30
step 10/42 - loss: 0.0384 - 3ms/step
step 20/42 - loss: 0.0366 - 3ms/step
step 30/42 - loss: 0.0352 - 3ms/step
step 40/42 - loss: 0.0603 - 3ms/step
step 42/42 - loss: 0.0136 - 3ms/step
Epoch 15/30
step 10/42 - loss: 0.0385 - 3ms/step
step 20/42 - loss: 0.0365 - 3ms/step
step 30/42 - loss: 0.0352 - 3ms/step
step 40/42 - loss: 0.0603 - 3ms/step
step 42/42 - loss: 0.0136 - 3ms/step
Epoch 16/30
step 10/42 - loss: 0.0386 - 3ms/step
step 20/42 - loss: 0.0365 - 3ms/step
step 30/42 - loss: 0.0352 - 3ms/step
step 40/42 - loss: 0.0603 - 3ms/step
step 42/42 - loss: 0.0137 - 3ms/step
Epoch 17/30
step 10/42 - loss: 0.0386 - 3ms/step
step 20/42 - loss: 0.0364 - 3ms/step
step 30/42 - loss: 0.0351 - 3ms/step
step 40/42 - loss: 0.0602 - 3ms/step
step 42/42 - loss: 0.0137 - 3ms/step
Epoch 18/30
step 10/42 - loss: 0.0387 - 3ms/step
step 20/42 - loss: 0.0364 - 3ms/step
step 30/42 - loss: 0.0351 - 3ms/step
step 40/42 - loss: 0.0602 - 3ms/step
step 42/42 - loss: 0.0137 - 3ms/step
Epoch 19/30
step 10/42 - loss: 0.0387 - 3ms/step
step 20/42 - loss: 0.0363 - 3ms/step
step 30/42 - loss: 0.0351 - 3ms/step
step 40/42 - loss: 0.0602 - 3ms/step
step 42/42 - loss: 0.0137 - 3ms/step
Epoch 20/30
step 10/42 - loss: 0.0388 - 3ms/step
step 20/42 - loss: 0.0362 - 3ms/step
step 30/42 - loss: 0.0350 - 3ms/step
step 40/42 - loss: 0.0601 - 3ms/step
step 42/42 - loss: 0.0137 - 3ms/step
Epoch 21/30
step 10/42 - loss: 0.0388 - 3ms/step
step 20/42 - loss: 0.0362 - 3ms/step
step 30/42 - loss: 0.0350 - 3ms/step
step 40/42 - loss: 0.0601 - 3ms/step
step 42/42 - loss: 0.0138 - 3ms/step
Epoch 22/30
step 10/42 - loss: 0.0389 - 3ms/step
step 20/42 - loss: 0.0361 - 3ms/step
step 30/42 - loss: 0.0350 - 3ms/step
step 40/42 - loss: 0.0600 - 3ms/step
step 42/42 - loss: 0.0138 - 3ms/step
Epoch 23/30
step 10/42 - loss: 0.0389 - 3ms/step
step 20/42 - loss: 0.0360 - 3ms/step
step 30/42 - loss: 0.0350 - 3ms/step
step 40/42 - loss: 0.0599 - 3ms/step
step 42/42 - loss: 0.0138 - 3ms/step
Epoch 24/30
step 10/42 - loss: 0.0389 - 3ms/step
step 20/42 - loss: 0.0360 - 3ms/step
step 30/42 - loss: 0.0349 - 3ms/step
step 40/42 - loss: 0.0599 - 3ms/step
step 42/42 - loss: 0.0138 - 3ms/step
Epoch 25/30
step 10/42 - loss: 0.0390 - 3ms/step
step 20/42 - loss: 0.0359 - 3ms/step
step 30/42 - loss: 0.0349 - 3ms/step
step 40/42 - loss: 0.0598 - 3ms/step
step 42/42 - loss: 0.0138 - 3ms/step
Epoch 26/30
step 10/42 - loss: 0.0390 - 3ms/step
step 20/42 - loss: 0.0358 - 3ms/step
step 30/42 - loss: 0.0349 - 3ms/step
step 40/42 - loss: 0.0597 - 3ms/step
step 42/42 - loss: 0.0139 - 3ms/step
Epoch 27/30
step 10/42 - loss: 0.0390 - 3ms/step
step 20/42 - loss: 0.0357 - 3ms/step
step 30/42 - loss: 0.0349 - 3ms/step
step 40/42 - loss: 0.0597 - 3ms/step
step 42/42 - loss: 0.0139 - 3ms/step
Epoch 28/30
step 10/42 - loss: 0.0391 - 3ms/step
step 20/42 - loss: 0.0357 - 3ms/step
step 30/42 - loss: 0.0349 - 3ms/step
step 40/42 - loss: 0.0596 - 3ms/step
step 42/42 - loss: 0.0139 - 3ms/step
Epoch 29/30
step 10/42 - loss: 0.0391 - 3ms/step
step 20/42 - loss: 0.0356 - 3ms/step
step 30/42 - loss: 0.0349 - 3ms/step
step 40/42 - loss: 0.0595 - 3ms/step
step 42/42 - loss: 0.0139 - 3ms/step
Epoch 30/30
step 10/42 - loss: 0.0391 - 3ms/step
step 20/42 - loss: 0.0355 - 3ms/step
step 30/42 - loss: 0.0348 - 3ms/step
step 40/42 - loss: 0.0595 - 3ms/step
step 42/42 - loss: 0.0140 - 3ms/step
Predict begin...
step 80/80 [==============================] - 2ms/step        
Predict samples: 80
test_dataset.x.shape (80, 12, 1)
test_dataset.y.shape (80, 1)

在这里插入图片描述

深耕AI
关注
  • 11
    点赞
  • 踩
  • 4
    收藏
    觉得还不错? 一键收藏
  • 打赏
    打赏
  • 0
    评论
python预测未来数据步骤_如何使用sklearnpython预测未来的数据帧？
weixin_33179060的博客
01-30 2370
您的代码使用此数据帧作为X来生成预测：df = df[['Adj. Close', 'HL_PCT', 'PCT_change', 'Adj. Volume']]这意味着，如果你想预测未来五年的价格，你需要这些['Adj. Close', 'HL_PCT', 'PCT_change', 'Adj. Volume']数据点来预测未来的价格。在请注意，您的图像中的预测是从历史数据中创建的，这些数据在这...
Paddle飞桨学习日记: LSTM股票预测
weixin_46754446的博客
06-24 2266
Paddle飞桨学习日记: LSTM股票预测本项目代码基于PaddleHub的开源代码。下面是项目完整代码预测结果 本次学习，打算初步通过LSTM来了解Paddle的一些细节 本项目代码基于PaddleHub的开源代码。 本文原始项目地址： https://aistudio.baidu.com/aistudio/projectdetail/174237 本文数据： https://aistudio.baidu.com/aistudio/datasetdetail/3580 下面是项目完整代码 #引用库文件
未来数据预测基于过去一段时间的数据序列，预测未来一天的数据（例如股票、温度等）
qq_70010893的博客
08-30 316
使用LSTM进行预测。
excel数据分析 - 3种预测数据趋势的技巧总结
Venassa佳的博客
10-07 1万+
一 预测工作表；二 预测函数TREND；三 散点图的趋势线
利用预测模型对所有数据下一年份的数据进行预测代码
m0_56659620的博客
08-14 487
【代码】利用预测模型对所有数据下一年份的数据进行预测代码。
用Python进行数据预测
很酷的站长的博客
09-09 5715
本文以Python进行数据预测为中心，从数据准备、特征工程、模型选择与训练、模型评估与优化等方面进行了详细的阐述。通过合理的数据处理和模型选择，结合Python强大的数据科学库，我们可以实现准确的数据预测和分析。希望本文对你在数据预测领域的学习和实践有所帮助。
Jupyter Notebook-基于Paddle多层LSTM模型的风电时序预测+源代码+文档说明
最新发布
08-11
基于Paddle多层LSTM模型的风电时序预测 - 不懂运行，下载完可以私聊问，可远程教学 该资源内项目源码是个人的毕设，代码都测试ok，都是运行成功后才上传资源，答辩评审平均分达到96分，放心下载使用！ 1、该资源内...
基于Paddle多层LSTM模型的风电时序预测.zip
05-22
LSTM (Long Short-Term Memory) 是一种特殊的循环神经网络（RNN）架构，用于处理...由于LSTM能够有效地处理长期依赖关系，它在许多序列建模任务中都取得了很好的效果，如语音识别、文本生成、机器翻译、时序预测等。
使用paddlepaddle的LSTM进行文本分类的代码
u010569893的博客
06-15 938
用paddlepaddle的LSTM进行文本分类的代码
paddle_RNN_LSTM(3).zip
12-31
标题中的"paddle_RNN_LSTM(3).zip"表明这是一个关于PaddlePaddle实现的LSTM模型的压缩包，可能包含了作者从头实现的单向、多层以及双向LSTM的代码，以及与PaddlePaddle内置LSTM模块的比较实验。 描述中提到，该资源...
python预测未来数据步骤_有没有简单的方法在python中推断数据点到未来？
weixin_32172693的博客
01-30 293
外推生成垃圾太容易了;试试这个.当然可以进行许多不同的推断;一些产生明显的垃圾,一些非明显的垃圾,许多是不明确的.""" extrapolate y,m,d data with scipy UnivariateSpline """import numpy as npfrom scipy.interpolate import UnivariateSpline# pydoc scipy.interpo...
如何使用Python进行时间序列分析与预测
程序员光剑
12-27 515
1.背景介绍 时间序列分析和预测是一种对时间顺序有序的观测数据进行分析和预测的方法。它广泛应用于各个领域，如金融、商业、天气、科学研究等。随着数据量的增加，人工智能和机器学习技术的发展为时间序列分析和预测提供了强大的支持。Python是一种流行的编程语言，拥有丰富的数据分析和机器学习库，使其成为时间序列分析和预测的理想工具。 在本文中，我们将讨论以下主题： 背景介绍 核心概念与联系 核心算...
用python爬取天气数据进行预测及可视化分析
xzz_777c的博客
03-21 3766
天气预报我们每天都会关注，我们可以根据未来的天气增减衣物、安排出行，每天的气温、风速风向、相对湿度、空气质量等成为关注的焦点。本次使用python中requests和BeautifulSoup库对中国天气网当天和未来14天的数据进行爬取，保存为csv文件，之后用matplotlib、numpy、pandas对数据进行可视化处理和分析，得到温湿度度变化曲线、空气质量图、风向雷达图等结果，为获得未来天气信息提供了有效方法。
Machine Learning
McGTSY3的专栏
08-17 363
前言： 　　深度学习话题十分火热，网上的资料也非常多，这的确很头疼，太容易迷失。个人认为寻找大牛的授课ppt作为入门方式就可以，跟随大牛的脚步先画出一条直线，再补充骨肉。Anyway,这篇文章十分适合机器学习初学者迅速了解深度学习，了解清楚什么是深度学习，如何做及现在最火的卷积神经网络。 教材链接：李宏毅 Let us rock！ 1.什么是机器学习 机器学习就是寻找一个函数。函数的输入是现实世界的信息，输出是一个抽象结果。 　　插入个人理解，可以跳过：每个函数的定义域、值域...
使用 Conv1D-LSTM 进行时间序列预测：预测多个未来时间步【优化】
早起CaiCai的学习空间
05-16 6218
Conv1D-LSTM时间序列预测
BP神经网络如何用历史数据预测未来数据
千枫的小白博客
05-13 3万+
本文主要为了解决如何用BP神经网络由历史的目标数据与因素数据去预测未来的目标数据。Bp神经网络的具体算法步骤与代码在网络上已经有很多大佬写过了，本文提供了将其应用于预测的具体实践方法。（附简单易懂python代码）
时间序列模型Prophet使用详细讲解
热门推荐
anshuai_aw1的博客
10-26 15万+
之前我们已经讲过了如何在Windows系统下安装Python版本的Prophet。详细见这里。 接下来的几个部分，我们说下如何使用Prophet，以此来体验下Prophet的丰富内容。内容会比较多，主要翻译自官方文档。教程中使用的数据集可在 Prophet 的 github 主页 中的 examples 文件夹 内下载得到。 目录 一、简易入门 二、饱和预测 2.1 预测饱和增长 2....
在Python中使用LSTM和PyTorch进行时间序列预测
大数据部落
10-29 1万+
原文链接：http://tecdat.cn/?p=8145 顾名思义，时间序列数据是一种随时间变化的数据类型。例如，24小时内的温度，一个月内各种产品的价格，一年中特定公司的股票价格。诸如长期短期记忆网络（LSTM）之类的高级深度学习模型能够捕获时间序列数据中的模式，因此可用于对数据的未来趋势进行预测。在本文中，您将看到如何使用LSTM算法使用时间序列数据进行将来的预测。 数据集......
基于paddle2.0实现知识蒸馏的代码
06-05
以下是一个基于PaddlePaddle 2.0实现知识蒸馏的代码示例： ```python import paddle import paddle.nn.functional as F from paddle.vision.models import resnet18 # 加载预训练的教师模型和数据集 teacher_model...
评论
添加红包

请填写红包祝福语或标题

个

红包个数最小为10个

元

红包金额最低5元

当前余额3.43元 前往充值 >
需支付：10.00元
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包

打赏作者

深耕AI

谢谢鼓励~我将继续创作优质博文

¥1 ¥2 ¥4 ¥6 ¥10 ¥20
扫码支付：¥1
获取中
扫码支付

您的余额不足，请更换扫码支付或充值

打赏作者

实付元
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明：

1.余额是钱包充值的虚拟货币，按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载，可以购买VIP、付费专栏及课程。

余额充值