yangwohenmai / LSTM

基于LSTM的时间序列预测研究
2.73k stars 696 forks source link

> > 你好,我想请教一下,我学习了一下LSTM多变量3这个完整的算例,结果看上去预测值会比实际值滞后一格,请问这是为什么。从代码上看标签都是取得数据重构后最后一列的数据,这个现象是正常的嘛。 ![image](https://user-images.githubusercontent.com/92704247/138576889-fb9428f6-cbf1-42b9-a1c9-0623ea8abce7.png) ![image](https://user-images.githubusercontent.com/92704247/138576898-b26fa9f4-0a1c-47da-a57a-9500fbf6f3a7.png) #5

Closed yangwohenmai closed 2 years ago

yangwohenmai commented 2 years ago

你好,我想请教一下,我学习了一下LSTM多变量3这个完整的算例,结果看上去预测值会比实际值滞后一格,请问这是为什么。从代码上看标签都是取得数据重构后最后一列的数据,这个现象是正常的嘛。 image image

是正常的,主要原因是这是这个网络只是一个用于学习的基本的简单网络,以及训练数据集并不足够,并没有训练得到很好的效果。如果想进一步提高准确率,需要对网络模型进行进一步的优化

补充多问一下,主要的运行过程中最后计算得到的RMSE是根据这个错位结果来的,导致RMSE看上去挺大的。别的算例里面有训练数据集不足的,确实是看上去是简单的平移,这个算例初始的训练集是一年,我把它增加到两年三年好像也没有改善,麻烦请教一下训练集大概要是多少才能不出现这种现象。 另外,真的很感谢,算例讲得其实很清楚,有很多注释,算是我神经网络的入门学习资料了!

Originally posted by @Zhujh0224 in https://github.com/yangwohenmai/LSTM/issues/4#issuecomment-951464750

yangwohenmai commented 2 years ago

仅仅去增加数据量,有用但也是不够的。要考虑从网络结构的调整,参数的优化,以及训练数据的特征工程入手。LSTM的本质是RNN,或多或少都会导致新数据比旧数据的权重大

1005848593 commented 2 years ago

image 我想请问这边把数据转化为有监督数据的作用是什么?采用转化之后的数据进行训练不会使数据泄露到模型中吗?谢谢!I would like to ask what is the purpose of transforming the data into supervised data over here? Won't the data leak into the model by using the transformed data for training? Thank you!

yangwohenmai commented 2 years ago

监督学习数据是训练时间序列模型时一种基本的输入格式,我们通过让机器看到历史的输入和未来的结果,从而学习到历史和未来之间的映射关系。这种转化并不会使数据泄露,我们避免数据泄露的方法分是割出 训练集 测试集 验证集,再分别对网络进行训练和数据验证。 supervised data is a general input format when training time series models. By letting the machine see historical input and future results, the machine can learn the mapping relationship between history and future. This transformation will not cause data leakage. The way to avoid data leakage is to divide the training set, test set and verification set, and then train and verify the network respectively

image 我想请问这边把数据转化为有监督数据的作用是什么?采用转化之后的数据进行训练不会使数据泄露到模型中吗?谢谢!I would like to ask what is the purpose of transforming the data into supervised data over here? Won't the data leak into the model by using the transformed data for training? Thank you!

1005848593 commented 2 years ago

非常感谢您的回信,请问把数据转化为有监督数据是属于数据预处理部分把。但是在数据预处理中怎么描述这一过程呢。谢谢您!Thank you very much for your reply. The transformation of data into supervised data is part of the data preprocessing. But how to describe this process in the data preprocessing. Thank you!

阔??阔、? @.***

 

------------------ 原始邮件 ------------------ 发件人: "yangwohenmai/LSTM" @.>; 发送时间: 2022年9月29日(星期四) 凌晨0:51 @.>; @.**@.>; 主题: Re: [yangwohenmai/LSTM] > > 你好,我想请教一下,我学习了一下LSTM多变量3这个完整的算例,结果看上去预测值会比实际值滞后一格,请问这是为什么。从代码上看标签都是取得数据重构后最后一列的数据,这个现象是正常的嘛。 image ![imag=?UTF-8?Q?e]=28https://user-images.g

监督学习数据是训练时间序列模型时一种基本的输入格式,我们通过让机器看到历史的输入和未来的结果,从而学习到历史和未来之间的映射关系。这种转化并不会使数据泄露,我们避免数据泄露的方法分是割出 训练集 测试集 验证集,再分别对网络进行训练和数据验证。 supervised data is a general input format when training time series models. By letting the machine see historical input and future results, the machine can learn the mapping relationship between history and future. This transformation will not cause data leakage. The way to avoid data leakage is to divide the training set, test set and verification set, and then train and verify the network respectively

我想请问这边把数据转化为有监督数据的作用是什么?采用转化之后的数据进行训练不会使数据泄露到模型中吗?谢谢!I would like to ask what is the purpose of transforming the data into supervised data over here? Won't the data leak into the model by using the transformed data for training? Thank you!

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>

yangwohenmai commented 2 years ago

非常感谢您的回信,请问把数据转化为有监督数据是属于数据预处理部分把。但是在数据预处理中怎么描述这一过程呢。谢谢您!Thank you very much for your reply. The transformation of data into supervised data is part of the data preprocessing. But how to describe this process in the data preprocessing. Thank you! 阔??阔、? @.   ------------------ 原始邮件 ------------------ 发件人: "yangwohenmai/LSTM" @.>; 发送时间: 2022年9月29日(星期四) 凌晨0:51 @.>; @*.**@*.>; 主题: Re: [yangwohenmai/LSTM] > > 你好,我想请教一下,我学习了一下LSTM多变量3这个完整的算例,结果看上去预测值会比实际值滞后一格,请问这是为什么。从代码上看标签都是取得数据重构后最后一列的数据,这个现象是正常的嘛。 image ![imag=?UTF-8?Q?e]=28https://user-images.g 监督学习数据是训练时间序列模型时一种基本的输入格式,我们通过让机器看到历史的输入和未来的结果,从而学习到历史和未来之间的映射关系。这种转化并不会使数据泄露,我们避免数据泄露的方法分是割出 训练集 测试集 验证集,再分别对网络进行训练和数据验证。 supervised data is a general input format when training time series models. By letting the machine see historical input and future results, the machine can learn the mapping relationship between history and future. This transformation will not cause data leakage. The way to avoid data leakage is to divide the training set, test set and verification set, and then train and verify the network respectively 我想请问这边把数据转化为有监督数据的作用是什么?采用转化之后的数据进行训练不会使数据泄露到模型中吗?谢谢!I would like to ask what is the purpose of transforming the data into supervised data over here? Won't the data leak into the model by using the transformed data for training? Thank you! — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.>

我不太清楚你说的描述指的是什么,但一般来说这属于“数据特征工程”的一部分。 I'm not sure what you mean that how to description, but generally it is part of the "data feature engineering".

1005848593 commented 2 years ago

非常感谢您的耐心回答,我想问的是将数据转化为有监督的数据这一步骤在模型训练过程中应该如何表达呢?谢谢! Thank you very much for your patience, I would like to ask how this step of transforming data into supervised data should be expressed in the model training process? Thanks!

阔??阔、? @.***

 

------------------ 原始邮件 ------------------ 发件人: "yangwohenmai/LSTM" @.>; 发送时间: 2022年9月29日(星期四) 中午1:33 @.>; @.**@.>; 主题: Re: [yangwohenmai/LSTM] > > 你好,我想请教一下,我学习了一下LSTM多变量3这个完整的算例,结果看上去预测值会比实际值滞后一格,请问这是为什么。从代码上看标签都是取得数据重构后最后一列的数据,这个现象是正常的嘛。 image ![imag=?UTF-8?Q?e]=28https://user-images.g

非常感谢您的回信,请问把数据转化为有监督数据是属于数据预处理部分把。但是在数据预处理中怎么描述这一过程呢。谢谢您!Thank you very much for your reply. The transformation of data into supervised data is part of the data preprocessing. But how to describe this process in the data preprocessing. Thank you! 阔??阔、? @.   … ------------------ 原始邮件 ------------------ 发件人: "yangwohenmai/LSTM" @.>; 发送时间: 2022年9月29日(星期四) 凌晨0:51 @.>; @.@.>; 主题: Re: [yangwohenmai/LSTM] > > 你好,我想请教一下,我学习了一下LSTM多变量3这个完整的算例,结果看上去预测值会比实际值滞后一格,请问这是为什么。从代码上看标签都是取得数据重构后最后一列的数据,这个现象是正常的嘛。 ![imag=?UTF-8?Q?e]=28https://user-images.g 监督学习数据是训练时间序列模型时一种基本的输入格式,我们通过让机器看到历史的输入和未来的结果,从而学习到历史和未来之间的映射关系。这种转化并不会使数据泄露,我们避免数据泄露的方法分是割出 训练集 测试集 验证集,再分别对网络进行训练和数据验证。 supervised data is a general input format when training time series models. By letting the machine see historical input and future results, the machine can learn the mapping relationship between history and future. This transformation will not cause data leakage. The way to avoid data leakage is to divide the training set, test set and verification set, and then train and verify the network respectively 我想请问这边把数据转化为有监督数据的作用是什么?采用转化之后的数据进行训练不会使数据泄露到模型中吗?谢谢!I would like to ask what is the purpose of transforming the data into supervised data over here? Won't the data leak into the model by using the transformed data for training? Thank you! — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.>

我不太清楚你说的描述指的是什么,但一般来说这属于“数据特征工程”的一部分。 I'm not sure what you mean that how to description, but generally it is part of the "data feature engineering".

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>

1005848593 commented 2 years ago

非要不好意思再次打扰到您,我想请问一下这边将测试集当作为验证集进行验证,这一步骤是不是存在数据泄露的情况?非常感谢! I'm sorry to bother you again, but I'd like to ask if there is any data leakage in this step when the test set is used as the validation set for verification? Thank you very much!

阔??阔、? @.***

 

------------------ 原始邮件 ------------------ 发件人: "yangwohenmai/LSTM" @.>; 发送时间: 2022年9月29日(星期四) 中午1:33 @.>; @.**@.>; 主题: Re: [yangwohenmai/LSTM] > > 你好,我想请教一下,我学习了一下LSTM多变量3这个完整的算例,结果看上去预测值会比实际值滞后一格,请问这是为什么。从代码上看标签都是取得数据重构后最后一列的数据,这个现象是正常的嘛。 image ![imag=?UTF-8?Q?e]=28https://user-images.g

非常感谢您的回信,请问把数据转化为有监督数据是属于数据预处理部分把。但是在数据预处理中怎么描述这一过程呢。谢谢您!Thank you very much for your reply. The transformation of data into supervised data is part of the data preprocessing. But how to describe this process in the data preprocessing. Thank you! 阔??阔、? @.   … ------------------ 原始邮件 ------------------ 发件人: "yangwohenmai/LSTM" @.>; 发送时间: 2022年9月29日(星期四) 凌晨0:51 @.>; @.@.>; 主题: Re: [yangwohenmai/LSTM] > > 你好,我想请教一下,我学习了一下LSTM多变量3这个完整的算例,结果看上去预测值会比实际值滞后一格,请问这是为什么。从代码上看标签都是取得数据重构后最后一列的数据,这个现象是正常的嘛。 ![imag=?UTF-8?Q?e]=28https://user-images.g 监督学习数据是训练时间序列模型时一种基本的输入格式,我们通过让机器看到历史的输入和未来的结果,从而学习到历史和未来之间的映射关系。这种转化并不会使数据泄露,我们避免数据泄露的方法分是割出 训练集 测试集 验证集,再分别对网络进行训练和数据验证。 supervised data is a general input format when training time series models. By letting the machine see historical input and future results, the machine can learn the mapping relationship between history and future. This transformation will not cause data leakage. The way to avoid data leakage is to divide the training set, test set and verification set, and then train and verify the network respectively 我想请问这边把数据转化为有监督数据的作用是什么?采用转化之后的数据进行训练不会使数据泄露到模型中吗?谢谢!I would like to ask what is the purpose of transforming the data into supervised data over here? Won't the data leak into the model by using the transformed data for training? Thank you! — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.>

我不太清楚你说的描述指的是什么,但一般来说这属于“数据特征工程”的一部分。 I'm not sure what you mean that how to description, but generally it is part of the "data feature engineering".

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>

yangwohenmai commented 2 years ago

非要不好意思再次打扰到您,我想请问一下这边将测试集当作为验证集进行验证,这一步骤是不是存在数据泄露的情况?非常感谢! I'm sorry to bother you again, but I'd like to ask if there is any data leakage in this step when the test set is used as the validation set for verification? Thank you very much! 阔??阔、? @.   ------------------ 原始邮件 ------------------ 发件人: "yangwohenmai/LSTM" @.>; 发送时间: 2022年9月29日(星期四) 中午1:33 @.>; @*.**@*.>; 主题: Re: [yangwohenmai/LSTM] > > 你好,我想请教一下,我学习了一下LSTM多变量3这个完整的算例,结果看上去预测值会比实际值滞后一格,请问这是为什么。从代码上看标签都是取得数据重构后最后一列的数据,这个现象是正常的嘛。 image ![imag=?UTF-8?Q?e]=28https://user-images.g 非常感谢您的回信,请问把数据转化为有监督数据是属于数据预处理部分把。但是在数据预处理中怎么描述这一过程呢。谢谢您!Thank you very much for your reply. The transformation of data into supervised data is part of the data preprocessing. But how to describe this process in the data preprocessing. Thank you! 阔??阔、? @.   … ------------------ 原始邮件 ------------------ 发件人: "yangwohenmai/LSTM" @.>; 发送时间: 2022年9月29日(星期四) 凌晨0:51 @.>; @.@.>; 主题: Re: [yangwohenmai/LSTM] > > 你好,我想请教一下,我学习了一下LSTM多变量3这个完整的算例,结果看上去预测值会比实际值滞后一格,请问这是为什么。从代码上看标签都是取得数据重构后最后一列的数据,这个现象是正常的嘛。 ![imag=?UTF-8?Q?e]=28https://user-images.g 监督学习数据是训练时间序列模型时一种基本的输入格式,我们通过让机器看到历史的输入和未来的结果,从而学习到历史和未来之间的映射关系。这种转化并不会使数据泄露,我们避免数据泄露的方法分是割出 训练集 测试集 验证集,再分别对网络进行训练和数据验证。 supervised data is a general input format when training time series models. By letting the machine see historical input and future results, the machine can learn the mapping relationship between history and future. This transformation will not cause data leakage. The way to avoid data leakage is to divide the training set, test set and verification set, and then train and verify the network respectively 我想请问这边把数据转化为有监督数据的作用是什么?采用转化之后的数据进行训练不会使数据泄露到模型中吗?谢谢!I would like to ask what is the purpose of transforming the data into supervised data over here? Won't the data leak into the model by using the transformed data for training? Thank you! — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.> 我不太清楚你说的描述指的是什么,但一般来说这属于“数据特征工程”的一部分。 I'm not sure what you mean that how to description, but generally it is part of the "data feature engineering". — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.>

测试集不能直接用来做验证集,如果我们有100条数据,我们会取80条做测试集,20条做验证集 The test set cannot be directly used as the verification set. If we have 100 pieces of data, we will take 80 pieces as the test set and 20 pieces as the verification set

1005848593 commented 2 years ago

但是您这边是用的测试集的数据做了验证集,如果按照您说的测试集和验证集分开的话这样会使模型效果很差,请问有什么方法能使模型预测精度更高?谢谢! But you have used the test set data for the validation set, if the test set and validation set are separated as you said, it will make the model poor. Thank you!

阔??阔、? @.***

 

------------------ 原始邮件 ------------------ 发件人: "yangwohenmai/LSTM" @.>; 发送时间: 2022年9月30日(星期五) 中午11:11 @.>; @.**@.>; 主题: Re: [yangwohenmai/LSTM] > > 你好,我想请教一下,我学习了一下LSTM多变量3这个完整的算例,结果看上去预测值会比实际值滞后一格,请问这是为什么。从代码上看标签都是取得数据重构后最后一列的数据,这个现象是正常的嘛。 image ![imag=?UTF-8?Q?e]=28https://user-images.g

非要不好意思再次打扰到您,我想请问一下这边将测试集当作为验证集进行验证,这一步骤是不是存在数据泄露的情况?非常感谢! I'm sorry to bother you again, but I'd like to ask if there is any data leakage in this step when the test set is used as the validation set for verification? Thank you very much! 阔??阔、? @.   … ------------------ 原始邮件 ------------------ 发件人: "yangwohenmai/LSTM" @.>; 发送时间: 2022年9月29日(星期四) 中午1:33 @.>; @.@.>; 主题: Re: [yangwohenmai/LSTM] > > 你好,我想请教一下,我学习了一下LSTM多变量3这个完整的算例,结果看上去预测值会比实际值滞后一格,请问这是为什么。从代码上看标签都是取得数据重构后最后一列的数据,这个现象是正常的嘛。 ![imag=?UTF-8?Q?e]=28https://user-images.g 非常感谢您的回信,请问把数据转化为有监督数据是属于数据预处理部分把。但是在数据预处理中怎么描述这一过程呢。谢谢您!Thank you very much for your reply. The transformation of data into supervised data is part of the data preprocessing. But how to describe this process in the data preprocessing. Thank you! 阔??阔、? @.   … ------------------ 原始邮件 ------------------ 发件人: "yangwohenmai/LSTM" @.>; 发送时间: 2022年9月29日(星期四) 凌晨0:51 @.>; @.@.>; 主题: Re: [yangwohenmai/LSTM] > > 你好,我想请教一下,我学习了一下LSTM多变量3这个完整的算例,结果看上去预测值会比实际值滞后一格,请问这是为什么。从代码上看标签都是取得数据重构后最后一列的数据,这个现象是正常的嘛。 ![imag=?UTF-8?Q?e]=28https://user-images.g 监督学习数据是训练时间序列模型时一种基本的输入格式,我们通过让机器看到历史的输入和未来的结果,从而学习到历史和未来之间的映射关系。这种转化并不会使数据泄露,我们避免数据泄露的方法分是割出 训练集 测试集 验证集,再分别对网络进行训练和数据验证。 supervised data is a general input format when training time series models. By letting the machine see historical input and future results, the machine can learn the mapping relationship between history and future. This transformation will not cause data leakage. The way to avoid data leakage is to divide the training set, test set and verification set, and then train and verify the network respectively 我想请问这边把数据转化为有监督数据的作用是什么?采用转化之后的数据进行训练不会使数据泄露到模型中吗?谢谢!I would like to ask what is the purpose of transforming the data into supervised data over here? Won't the data leak into the model by using the transformed data for training? Thank you! — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.> 我不太清楚你说的描述指的是什么,但一般来说这属于“数据特征工程”的一部分。 I'm not sure what you mean that how to description, but generally it is part of the "data feature engineering". — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.>

测试集不能直接用来做验证集,如果我们有100条数据,我们会取80条做测试集,20条做验证集 The test set cannot be directly used as the verification set. If we have 100 pieces of data, we will take 80 pieces as the test set and 20 pieces as the verification set

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>

yangwohenmai commented 2 years ago

但是您这边是用的测试集的数据做了验证集,如果按照您说的测试集和验证集分开的话这样会使模型效果很差,请问有什么方法能使模型预测精度更高?谢谢! But you have used the test set data for the validation set, if the test set and validation set are separated as you said, it will make the model poor. Thank you! 阔??阔、? @.   ------------------ 原始邮件 ------------------ 发件人: "yangwohenmai/LSTM" @.>; 发送时间: 2022年9月30日(星期五) 中午11:11 @.>; @*.**@*.>; 主题: Re: [yangwohenmai/LSTM] > > 你好,我想请教一下,我学习了一下LSTM多变量3这个完整的算例,结果看上去预测值会比实际值滞后一格,请问这是为什么。从代码上看标签都是取得数据重构后最后一列的数据,这个现象是正常的嘛。 image ![imag=?UTF-8?Q?e]=28https://user-images.g 非要不好意思再次打扰到您,我想请问一下这边将测试集当作为验证集进行验证,这一步骤是不是存在数据泄露的情况?非常感谢! I'm sorry to bother you again, but I'd like to ask if there is any data leakage in this step when the test set is used as the validation set for verification? Thank you very much! 阔??阔、? @.   … ------------------ 原始邮件 ------------------ 发件人: "yangwohenmai/LSTM" @.>; 发送时间: 2022年9月29日(星期四) 中午1:33 @.>; @.@.>; 主题: Re: [yangwohenmai/LSTM] > > 你好,我想请教一下,我学习了一下LSTM多变量3这个完整的算例,结果看上去预测值会比实际值滞后一格,请问这是为什么。从代码上看标签都是取得数据重构后最后一列的数据,这个现象是正常的嘛。 ![imag=?UTF-8?Q?e]=28https://user-images.g 非常感谢您的回信,请问把数据转化为有监督数据是属于数据预处理部分把。但是在数据预处理中怎么描述这一过程呢。谢谢您!Thank you very much for your reply. The transformation of data into supervised data is part of the data preprocessing. But how to describe this process in the data preprocessing. Thank you! 阔??阔、? @.   … ------------------ 原始邮件 ------------------ 发件人: "yangwohenmai/LSTM" @.>; 发送时间: 2022年9月29日(星期四) 凌晨0:51 @.>; @.@.>; 主题: Re: [yangwohenmai/LSTM] > > 你好,我想请教一下,我学习了一下LSTM多变量3这个完整的算例,结果看上去预测值会比实际值滞后一格,请问这是为什么。从代码上看标签都是取得数据重构后最后一列的数据,这个现象是正常的嘛。 ![imag=?UTF-8?Q?e]=28https://user-images.g 监督学习数据是训练时间序列模型时一种基本的输入格式,我们通过让机器看到历史的输入和未来的结果,从而学习到历史和未来之间的映射关系。这种转化并不会使数据泄露,我们避免数据泄露的方法分是割出 训练集 测试集 验证集,再分别对网络进行训练和数据验证。 supervised data is a general input format when training time series models. By letting the machine see historical input and future results, the machine can learn the mapping relationship between history and future. This transformation will not cause data leakage. The way to avoid data leakage is to divide the training set, test set and verification set, and then train and verify the network respectively 我想请问这边把数据转化为有监督数据的作用是什么?采用转化之后的数据进行训练不会使数据泄露到模型中吗?谢谢!I would like to ask what is the purpose of transforming the data into supervised data over here? Won't the data leak into the model by using the transformed data for training? Thank you! — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.> 我不太清楚你说的描述指的是什么,但一般来说这属于“数据特征工程”的一部分。 I'm not sure what you mean that how to description, but generally it is part of the "data feature engineering". — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.> 测试集不能直接用来做验证集,如果我们有100条数据,我们会取80条做测试集,20条做验证集 The test set cannot be directly used as the verification set. If we have 100 pieces of data, we will take 80 pieces as the test set and 20 pieces as the verification set — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.>

是哪部分代码?提升准确度的方法有很多,调整模型参数,使用更复杂的模型等 What part of the code is it? There are many methods to improve accuracy, such as adjusting model parameters, using more complex models, and so on

1005848593 commented 2 years ago

您看这部分验证集采用的test_x的数据,后面模型预测也是用的test_x的数据。当我把验证集的数据不用test_x时,模型效果就会很差。 You can see that this part of the validation set uses test_x data, and the model prediction later also uses test_x data. When I don't use the test_x data in the validation set, the model will be very poor.

阔??阔、? @.***

 

------------------ 原始邮件 ------------------ 发件人: "yangwohenmai/LSTM" @.>; 发送时间: 2022年9月30日(星期五) 中午11:18 @.>; @.**@.>; 主题: Re: [yangwohenmai/LSTM] > > 你好,我想请教一下,我学习了一下LSTM多变量3这个完整的算例,结果看上去预测值会比实际值滞后一格,请问这是为什么。从代码上看标签都是取得数据重构后最后一列的数据,这个现象是正常的嘛。 image ![imag=?UTF-8?Q?e]=28https://user-images.g

但是您这边是用的测试集的数据做了验证集,如果按照您说的测试集和验证集分开的话这样会使模型效果很差,请问有什么方法能使模型预测精度更高?谢谢! But you have used the test set data for the validation set, if the test set and validation set are separated as you said, it will make the model poor. Thank you! 阔??阔、? @.   … ------------------ 原始邮件 ------------------ 发件人: "yangwohenmai/LSTM" @.>; 发送时间: 2022年9月30日(星期五) 中午11:11 @.>; @.@.>; 主题: Re: [yangwohenmai/LSTM] > > 你好,我想请教一下,我学习了一下LSTM多变量3这个完整的算例,结果看上去预测值会比实际值滞后一格,请问这是为什么。从代码上看标签都是取得数据重构后最后一列的数据,这个现象是正常的嘛。 ![imag=?UTF-8?Q?e]=28https://user-images.g 非要不好意思再次打扰到您,我想请问一下这边将测试集当作为验证集进行验证,这一步骤是不是存在数据泄露的情况?非常感谢! I'm sorry to bother you again, but I'd like to ask if there is any data leakage in this step when the test set is used as the validation set for verification? Thank you very much! 阔??阔、? @.   … ------------------ 原始邮件 ------------------ 发件人: "yangwohenmai/LSTM" @.>; 发送时间: 2022年9月29日(星期四) 中午1:33 @.>; @.@.>; 主题: Re: [yangwohenmai/LSTM] > > 你好,我想请教一下,我学习了一下LSTM多变量3这个完整的算例,结果看上去预测值会比实际值滞后一格,请问这是为什么。从代码上看标签都是取得数据重构后最后一列的数据,这个现象是正常的嘛。 ![imag=?UTF-8?Q?e]=28https://user-images.g 非常感谢您的回信,请问把数据转化为有监督数据是属于数据预处理部分把。但是在数据预处理中怎么描述这一过程呢。谢谢您!Thank you very much for your reply. The transformation of data into supervised data is part of the data preprocessing. But how to describe this process in the data preprocessing. Thank you! 阔??阔、? @.   … ------------------ 原始邮件 ------------------ 发件人: "yangwohenmai/LSTM" @.>; 发送时间: 2022年9月29日(星期四) 凌晨0:51 @.>; @.@.>; 主题: Re: [yangwohenmai/LSTM] > > 你好,我想请教一下,我学习了一下LSTM多变量3这个完整的算例,结果看上去预测值会比实际值滞后一格,请问这是为什么。从代码上看标签都是取得数据重构后最后一列的数据,这个现象是正常的嘛。 ![imag=?UTF-8?Q?e]=28https://user-images.g 监督学习数据是训练时间序列模型时一种基本的输入格式,我们通过让机器看到历史的输入和未来的结果,从而学习到历史和未来之间的映射关系。这种转化并不会使数据泄露,我们避免数据泄露的方法分是割出 训练集 测试集 验证集,再分别对网络进行训练和数据验证。 supervised data is a general input format when training time series models. By letting the machine see historical input and future results, the machine can learn the mapping relationship between history and future. This transformation will not cause data leakage. The way to avoid data leakage is to divide the training set, test set and verification set, and then train and verify the network respectively 我想请问这边把数据转化为有监督数据的作用是什么?采用转化之后的数据进行训练不会使数据泄露到模型中吗?谢谢!I would like to ask what is the purpose of transforming the data into supervised data over here? Won't the data leak into the model by using the transformed data for training? Thank you! — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.> 我不太清楚你说的描述指的是什么,但一般来说这属于“数据特征工程”的一部分。 I'm not sure what you mean that how to description, but generally it is part of the "data feature engineering". — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.> 测试集不能直接用来做验证集,如果我们有100条数据,我们会取80条做测试集,20条做验证集 The test set cannot be directly used as the verification set. If we have 100 pieces of data, we will take 80 pieces as the test set and 20 pieces as the verification set — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.>

是哪部分代码?提升准确度的方法有很多,调整模型参数,使用更复杂的模型等 What part of the code is it? There are many methods to improve accuracy, such as adjusting model parameters, using more complex models, and so on

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>

1005848593 commented 2 years ago

请问是否可以跟您加一个联系方式,qq或者微信,这样方便我们交流。谢谢! qq:1005848593 微信:l18971054882 May I ask if I can add a contact with you, qq or WeChat, so that it is convenient for us to communicate. Thank you! qq:1005848593 WeChat:l18971054882

阔??阔、? @.***

 

------------------ 原始邮件 ------------------ 发件人: "yangwohenmai/LSTM" @.>; 发送时间: 2022年9月30日(星期五) 中午11:18 @.>; @.**@.>; 主题: Re: [yangwohenmai/LSTM] > > 你好,我想请教一下,我学习了一下LSTM多变量3这个完整的算例,结果看上去预测值会比实际值滞后一格,请问这是为什么。从代码上看标签都是取得数据重构后最后一列的数据,这个现象是正常的嘛。 image ![imag=?UTF-8?Q?e]=28https://user-images.g

但是您这边是用的测试集的数据做了验证集,如果按照您说的测试集和验证集分开的话这样会使模型效果很差,请问有什么方法能使模型预测精度更高?谢谢! But you have used the test set data for the validation set, if the test set and validation set are separated as you said, it will make the model poor. Thank you! 阔??阔、? @.   … ------------------ 原始邮件 ------------------ 发件人: "yangwohenmai/LSTM" @.>; 发送时间: 2022年9月30日(星期五) 中午11:11 @.>; @.@.>; 主题: Re: [yangwohenmai/LSTM] > > 你好,我想请教一下,我学习了一下LSTM多变量3这个完整的算例,结果看上去预测值会比实际值滞后一格,请问这是为什么。从代码上看标签都是取得数据重构后最后一列的数据,这个现象是正常的嘛。 ![imag=?UTF-8?Q?e]=28https://user-images.g 非要不好意思再次打扰到您,我想请问一下这边将测试集当作为验证集进行验证,这一步骤是不是存在数据泄露的情况?非常感谢! I'm sorry to bother you again, but I'd like to ask if there is any data leakage in this step when the test set is used as the validation set for verification? Thank you very much! 阔??阔、? @.   … ------------------ 原始邮件 ------------------ 发件人: "yangwohenmai/LSTM" @.>; 发送时间: 2022年9月29日(星期四) 中午1:33 @.>; @.@.>; 主题: Re: [yangwohenmai/LSTM] > > 你好,我想请教一下,我学习了一下LSTM多变量3这个完整的算例,结果看上去预测值会比实际值滞后一格,请问这是为什么。从代码上看标签都是取得数据重构后最后一列的数据,这个现象是正常的嘛。 ![imag=?UTF-8?Q?e]=28https://user-images.g 非常感谢您的回信,请问把数据转化为有监督数据是属于数据预处理部分把。但是在数据预处理中怎么描述这一过程呢。谢谢您!Thank you very much for your reply. The transformation of data into supervised data is part of the data preprocessing. But how to describe this process in the data preprocessing. Thank you! 阔??阔、? @.   … ------------------ 原始邮件 ------------------ 发件人: "yangwohenmai/LSTM" @.>; 发送时间: 2022年9月29日(星期四) 凌晨0:51 @.>; @.@.>; 主题: Re: [yangwohenmai/LSTM] > > 你好,我想请教一下,我学习了一下LSTM多变量3这个完整的算例,结果看上去预测值会比实际值滞后一格,请问这是为什么。从代码上看标签都是取得数据重构后最后一列的数据,这个现象是正常的嘛。 ![imag=?UTF-8?Q?e]=28https://user-images.g 监督学习数据是训练时间序列模型时一种基本的输入格式,我们通过让机器看到历史的输入和未来的结果,从而学习到历史和未来之间的映射关系。这种转化并不会使数据泄露,我们避免数据泄露的方法分是割出 训练集 测试集 验证集,再分别对网络进行训练和数据验证。 supervised data is a general input format when training time series models. By letting the machine see historical input and future results, the machine can learn the mapping relationship between history and future. This transformation will not cause data leakage. The way to avoid data leakage is to divide the training set, test set and verification set, and then train and verify the network respectively 我想请问这边把数据转化为有监督数据的作用是什么?采用转化之后的数据进行训练不会使数据泄露到模型中吗?谢谢!I would like to ask what is the purpose of transforming the data into supervised data over here? Won't the data leak into the model by using the transformed data for training? Thank you! — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.> 我不太清楚你说的描述指的是什么,但一般来说这属于“数据特征工程”的一部分。 I'm not sure what you mean that how to description, but generally it is part of the "data feature engineering". — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.> 测试集不能直接用来做验证集,如果我们有100条数据,我们会取80条做测试集,20条做验证集 The test set cannot be directly used as the verification set. If we have 100 pieces of data, we will take 80 pieces as the test set and 20 pieces as the verification set — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.>

是哪部分代码?提升准确度的方法有很多,调整模型参数,使用更复杂的模型等 What part of the code is it? There are many methods to improve accuracy, such as adjusting model parameters, using more complex models, and so on

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>

yangwohenmai commented 2 years ago

请问是否可以跟您加一个联系方式,qq或者微信,这样方便我们交流。谢谢! qq:1005848593 微信:l18971054882 May I ask if I can add a contact with you, qq or WeChat, so that it is convenient for us to communicate. Thank you! qq:1005848593 WeChat:l18971054882 阔??阔、? @.   ------------------ 原始邮件 ------------------ 发件人: "yangwohenmai/LSTM" @.>; 发送时间: 2022年9月30日(星期五) 中午11:18 @.>; @*.**@*.>; 主题: Re: [yangwohenmai/LSTM] > > 你好,我想请教一下,我学习了一下LSTM多变量3这个完整的算例,结果看上去预测值会比实际值滞后一格,请问这是为什么。从代码上看标签都是取得数据重构后最后一列的数据,这个现象是正常的嘛。 image ![imag=?UTF-8?Q?e]=28https://user-images.g 但是您这边是用的测试集的数据做了验证集,如果按照您说的测试集和验证集分开的话这样会使模型效果很差,请问有什么方法能使模型预测精度更高?谢谢! But you have used the test set data for the validation set, if the test set and validation set are separated as you said, it will make the model poor. Thank you! 阔??阔、? @.   … ------------------ 原始邮件 ------------------ 发件人: "yangwohenmai/LSTM" @.>; 发送时间: 2022年9月30日(星期五) 中午11:11 @.>; @.@.>; 主题: Re: [yangwohenmai/LSTM] > > 你好,我想请教一下,我学习了一下LSTM多变量3这个完整的算例,结果看上去预测值会比实际值滞后一格,请问这是为什么。从代码上看标签都是取得数据重构后最后一列的数据,这个现象是正常的嘛。 ![imag=?UTF-8?Q?e]=28https://user-images.g 非要不好意思再次打扰到您,我想请问一下这边将测试集当作为验证集进行验证,这一步骤是不是存在数据泄露的情况?非常感谢! I'm sorry to bother you again, but I'd like to ask if there is any data leakage in this step when the test set is used as the validation set for verification? Thank you very much! 阔??阔、? @.   … ------------------ 原始邮件 ------------------ 发件人: "yangwohenmai/LSTM" @.>; 发送时间: 2022年9月29日(星期四) 中午1:33 @.>; @.@.>; 主题: Re: [yangwohenmai/LSTM] > > 你好,我想请教一下,我学习了一下LSTM多变量3这个完整的算例,结果看上去预测值会比实际值滞后一格,请问这是为什么。从代码上看标签都是取得数据重构后最后一列的数据,这个现象是正常的嘛。 ![imag=?UTF-8?Q?e]=28https://user-images.g 非常感谢您的回信,请问把数据转化为有监督数据是属于数据预处理部分把。但是在数据预处理中怎么描述这一过程呢。谢谢您!Thank you very much for your reply. The transformation of data into supervised data is part of the data preprocessing. But how to describe this process in the data preprocessing. Thank you! 阔??阔、? @.   … ------------------ 原始邮件 ------------------ 发件人: "yangwohenmai/LSTM" @.>; 发送时间: 2022年9月29日(星期四) 凌晨0:51 @.>; @.@.>; 主题: Re: [yangwohenmai/LSTM] > > 你好,我想请教一下,我学习了一下LSTM多变量3这个完整的算例,结果看上去预测值会比实际值滞后一格,请问这是为什么。从代码上看标签都是取得数据重构后最后一列的数据,这个现象是正常的嘛。 ![imag=?UTF-8?Q?e]=28https://user-images.g 监督学习数据是训练时间序列模型时一种基本的输入格式,我们通过让机器看到历史的输入和未来的结果,从而学习到历史和未来之间的映射关系。这种转化并不会使数据泄露,我们避免数据泄露的方法分是割出 训练集 测试集 验证集,再分别对网络进行训练和数据验证。 supervised data is a general input format when training time series models. By letting the machine see historical input and future results, the machine can learn the mapping relationship between history and future. This transformation will not cause data leakage. The way to avoid data leakage is to divide the training set, test set and verification set, and then train and verify the network respectively 我想请问这边把数据转化为有监督数据的作用是什么?采用转化之后的数据进行训练不会使数据泄露到模型中吗?谢谢!I would like to ask what is the purpose of transforming the data into supervised data over here? Won't the data leak into the model by using the transformed data for training? Thank you! — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.> 我不太清楚你说的描述指的是什么,但一般来说这属于“数据特征工程”的一部分。 I'm not sure what you mean that how to description, but generally it is part of the "data feature engineering". — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.> 测试集不能直接用来做验证集,如果我们有100条数据,我们会取80条做测试集,20条做验证集 The test set cannot be directly used as the verification set. If we have 100 pieces of data, we will take 80 pieces as the test set and 20 pieces as the verification set — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.> 是哪部分代码?提升准确度的方法有很多,调整模型参数,使用更复杂的模型等 What part of the code is it? There are many methods to improve accuracy, such as adjusting model parameters, using more complex models, and so on — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.>

ok,I'll add you