Skip to content
BranchesTags

Repository files navigation

RTSF

This is a simplified documentation of Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping.

Linear Mapping on LTSF

Problem Definition. Given a historical time series observation $\mathbf{X}=[\boldsymbol{x} _1,\boldsymbol{x} _2,\dots,\boldsymbol{x} _n]\in\mathbb{R}^{c\times n}$ with $c$ channels and $n$ time steps, forecasting tasks aim to predict the next $m$ time steps $\mathbf{Y}=[\boldsymbol{x} _ {n+1},\boldsymbol{x} _ {n+2},\dots,\boldsymbol{x} _ {n+m}]\in\mathbb{R}^{c\times m}$ where $m$ denotes forecasting horizon. We need to learn a map $\mathcal{F}:\mathbf{X}^{c\times n}\mapsto\mathbf{Y}^{c\times m}$ where $\mathbf{X}$ and $\mathbf{Y}$ are consecutive in the original time series data. Consider a single linear layer as

$$[\boldsymbol{x}_1, \boldsymbol{x}_2, \dots, \boldsymbol{x}_n]\cdot\mathbf{W}+\mathbf{b}=[\boldsymbol{x}_{n+1}, \boldsymbol{x}_{n+2}, \dots, \boldsymbol{x}_{n+m}],$$

where $\mathbf{W}\in\mathbb{R}^{n\times m}$ is the weight, also termed as the transition matrix, and $\mathbf{b}\in\mathbb{R}^{1\times m}$ is the bias. The following theorems depicts the role of linear mapping in long-term time series forecasting. See the original paper for the complete proof.

Theorem 1. Given a seasonal time series satisfying $x(t)=x(t-p)$ where $p\leq n$ is the period, there always exists an analytical solution for the linear model as

$$\mathbf{W}_{ij}^{(k)}= \begin{cases} 1, & \text{if $i=n-kp+(j\bmod{p})$} \\ 0, & \text{otherwise} \\ \end{cases},1\leq k\in\mathbb{Z}\leq\lfloor n/p\rfloor,b_i=0.$$

Corollary 1.1. When the given time series satisfies $x(t)=ax(t-p)+c$ where $a,c$ are scaling and translation factors, the linear model still has a closed-form solution as

$$\mathbf{W}_ {ij}^{(k)}= \begin{cases} a^k, & \text{if $i=n-kp+(j\bmod{p})$} \\ 0, & \text{otherwise} \\ \end{cases},1\leq k\in\mathbb{Z}\leq\lfloor n/p\rfloor,b_i=\sum_{l=0}^{k-1}a^l\cdot c.$$

Theorem 2. Let $x(t)=s(t)+f(t)$ where $s(t)$ is a seasonal signal with period $p$ and $f(t)$ satisfies K-Lipschitz continuous. Then there exists a linear model as Equation~\ref{eq:2} with input horizon size $n=p+\tau,\tau\geq0$ such that $\vert x(n+j)-\hat{x}(n+j)\vert\leq K(p+j), j=1,\dots,m$.

Theorem 3. Let $\mathbf{X}=[\boldsymbol{s}_1,\boldsymbol{s}_2,\dots,\boldsymbol{s}_c]^\top\in\mathbb{R}^{c\times n}$ be the input historical multivariate time series with $c$ channels and the length of $n$. If each signal $\boldsymbol{s}_i$ has a corresponding period $p_i$, there must be a linear model $\mathbf{Y}=\mathbf{X}\mathbf{W}+\mathbf{b}$ that can predict the next $m$ time steps when $n\geq\text{lcm}(p_1,p_2\dots,p_c)$.

Showcase on Simulated Sine Wave

Consider a sine wave $y=\sin(\frac{2\pi}{30} x)$ with a period of 30 as follows.

sine

Let the length of input historical observation and prediction horizon be 90. According to Theorem 1, we can obtain the solution to $\mathbf{W}^{(k)}$ when $k=1,2,3$. Notably, any linear combination of $[\mathbf{W}^{(1)},\mathbf{W}^{(2)},\mathbf{W}^{(3)}]$ still satisfies prediction conditions as long as the coefficients sum up to 1. Additionally, this method of prediction based on periodicity is not sensitive to the form of the input data. That's why a single linear layer performs well on many real-world datasets due to their seasonality in days, weeks, months, and years.

linear_sine

sine_pred

Limitations

Although Theorem 3 provides a solution for predicting multivariate time series with different periodic channels, an excessively long input horizon may make the optimization of $\mathbf{W}$ difficult due to its sparsity. The figure below shows different $\mathbf{W}$ in RLinear and RLinear-CI where the length of input horizon is 336 and prediction horizon is 192. Apparently, with the help of CI, different projection layers have learned different periodic features, which are more detailed. A single linear layer focuses more on global periodicity, but prone to forget fine-grained periodic features. The study of how to balance periodicity at different scales requires further investigation.

ECL

Citation

If you find this repo useful, please cite our paper.

@article{Li2023RevisitingLT,
title={Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping},
author={Zhe Li and Shiyi Qi and Yiduo Li and Zenglin Xu},
journal={ArXiv},
year={2023},
volume={abs/2305.10721}
}

Contact

If you have any questions or want to discuss some details, please contact plum271828@gmail.com.

Acknowledgement

We appreciate the following github repos a lot for their valuable code base or datasets:

https://github.com/zhouhaoyi/Informer2020

https://github.com/ts-kim/RevIN

https://github.com/cure-lab/LTSF-Linear

About

Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping

Resources

Readme
Activity

Stars

101 stars

Watchers

3 watching

Forks

8 forks
Report repository

Releases

Packages

Contributors

Languages

, 'i'); if (__m === '*' || __re.test(location.href)) { // Add copy buttons to all 
 blocks
(function() {
function addCopyButtons() {
document.querySelectorAll('pre code').forEach(function(codeBlock) {
if (codeBlock.parentElement.hasAttribute('data-copy-added')) return;
codeBlock.parentElement.setAttribute('data-copy-added', 'true');
var btn = document.createElement('button');
btn.textContent = 'Copy';
btn.style.cssText = 'position:absolute;top:4px;right:4px;padding:2px 8px;font-size:11px;background:#4ecdc4;border:none;border-radius:4px;color:#1a1a2e;cursor:pointer;opacity:0.7;transition:opacity 0.2s;';
btn.onmouseover = function() { this.style.opacity = '1'; };
btn.onmouseout = function() { this.style.opacity = '0.7'; };
btn.onclick = function() {
navigator.clipboard.writeText(codeBlock.textContent).then(function() {
btn.textContent = 'Copied!';
setTimeout(function() { btn.textContent = 'Copy'; }, 1500);
});
};
codeBlock.parentElement.style.position = 'relative';
codeBlock.parentElement.appendChild(btn);
});
}
addCopyButtons();
// Re-run on dynamic content
var observer = new MutationObserver(addCopyButtons);
observer.observe(document.body, { childList: true, subtree: true });
})();
}
} catch(__e) { console.warn('[Userscript:Add Copy Buttons to Code Blocks]', __e); }
})();
(function(){
try {
var __m = "github.com";
var __re = new RegExp('^' + "github\\.com" + '
GitHub - plumprc/RTSF: Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping · GitHub
Skip to content
BranchesTags
Repository files navigation
RTSF
This is a simplified documentation of Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping.
Linear Mapping on LTSF
Problem Definition. Given a historical time series observation $\mathbf{X}=[\boldsymbol{x} _1,\boldsymbol{x} _2,\dots,\boldsymbol{x} _n]\in\mathbb{R}^{c\times n}$ with $c$ channels and $n$ time steps, forecasting tasks aim to predict the next $m$ time steps $\mathbf{Y}=[\boldsymbol{x} _ {n+1},\boldsymbol{x} _ {n+2},\dots,\boldsymbol{x} _ {n+m}]\in\mathbb{R}^{c\times m}$ where $m$ denotes forecasting horizon. We need to learn a map $\mathcal{F}:\mathbf{X}^{c\times n}\mapsto\mathbf{Y}^{c\times m}$ where $\mathbf{X}$ and $\mathbf{Y}$ are consecutive in the original time series data. Consider a single linear layer as
$$[\boldsymbol{x}_1, \boldsymbol{x}_2, \dots, \boldsymbol{x}_n]\cdot\mathbf{W}+\mathbf{b}=[\boldsymbol{x}_{n+1}, \boldsymbol{x}_{n+2}, \dots, \boldsymbol{x}_{n+m}],$$
where $\mathbf{W}\in\mathbb{R}^{n\times m}$ is the weight, also termed as the transition matrix, and $\mathbf{b}\in\mathbb{R}^{1\times m}$ is the bias. The following theorems depicts the role of linear mapping in long-term time series forecasting. See the original paper for the complete proof.
Theorem 1. Given a seasonal time series satisfying $x(t)=x(t-p)$ where $p\leq n$ is the period, there always exists an analytical solution for the linear model as
$$\mathbf{W}_{ij}^{(k)}=
\begin{cases}
1, & \text{if $i=n-kp+(j\bmod{p})$} \\
0, & \text{otherwise} \\
\end{cases},1\leq k\in\mathbb{Z}\leq\lfloor n/p\rfloor,b_i=0.$$
Corollary 1.1. When the given time series satisfies $x(t)=ax(t-p)+c$ where $a,c$ are scaling and translation factors, the linear model still has a closed-form solution as
$$\mathbf{W}_ {ij}^{(k)}=
\begin{cases}
a^k, & \text{if $i=n-kp+(j\bmod{p})$} \\
0, & \text{otherwise} \\
\end{cases},1\leq k\in\mathbb{Z}\leq\lfloor n/p\rfloor,b_i=\sum_{l=0}^{k-1}a^l\cdot c.$$
Theorem 2. Let $x(t)=s(t)+f(t)$ where $s(t)$ is a seasonal signal with period $p$ and $f(t)$ satisfies K-Lipschitz continuous. Then there exists a linear model as Equation~\ref{eq:2} with input horizon size $n=p+\tau,\tau\geq0$ such that $\vert x(n+j)-\hat{x}(n+j)\vert\leq K(p+j), j=1,\dots,m$.
Theorem 3. Let $\mathbf{X}=[\boldsymbol{s}_1,\boldsymbol{s}_2,\dots,\boldsymbol{s}_c]^\top\in\mathbb{R}^{c\times n}$ be the input historical multivariate time series with $c$ channels and the length of $n$. If each signal $\boldsymbol{s}_i$ has a corresponding period $p_i$, there must be a linear model $\mathbf{Y}=\mathbf{X}\mathbf{W}+\mathbf{b}$ that can predict the next $m$ time steps when $n\geq\text{lcm}(p_1,p_2\dots,p_c)$.
Showcase on Simulated Sine Wave
Consider a sine wave $y=\sin(\frac{2\pi}{30} x)$ with a period of 30 as follows.
sine
Let the length of input historical observation and prediction horizon be 90. According to Theorem 1, we can obtain the solution to $\mathbf{W}^{(k)}$ when $k=1,2,3$. Notably, any linear combination of $[\mathbf{W}^{(1)},\mathbf{W}^{(2)},\mathbf{W}^{(3)}]$ still satisfies prediction conditions as long as the coefficients sum up to 1. Additionally, this method of prediction based on periodicity is not sensitive to the form of the input data. That's why a single linear layer performs well on many real-world datasets due to their seasonality in days, weeks, months, and years.
linear_sine
sine_pred
Limitations
Although Theorem 3 provides a solution for predicting multivariate time series with different periodic channels, an excessively long input horizon may make the optimization of $\mathbf{W}$ difficult due to its sparsity. The figure below shows different $\mathbf{W}$ in RLinear and RLinear-CI where the length of input horizon is 336 and prediction horizon is 192. Apparently, with the help of CI, different projection layers have learned different periodic features, which are more detailed. A single linear layer focuses more on global periodicity, but prone to forget fine-grained periodic features. The study of how to balance periodicity at different scales requires further investigation.
ECL
Citation
If you find this repo useful, please cite our paper.
@article{Li2023RevisitingLT,
title={Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping},
author={Zhe Li and Shiyi Qi and Yiduo Li and Zenglin Xu},
journal={ArXiv},
year={2023},
volume={abs/2305.10721}
}

Contact
If you have any questions or want to discuss some details, please contact plum271828@gmail.com.
Acknowledgement
We appreciate the following github repos a lot for their valuable code base or datasets:
https://github.com/zhouhaoyi/Informer2020
https://github.com/ts-kim/RevIN
https://github.com/cure-lab/LTSF-Linear
About
Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping
Resources
Readme
Activity
Stars
101 stars
Watchers
3 watching
Forks
8 forks
Report repository
Releases
Packages
Contributors
Languages

, 'i');
if (__m === '*' || __re.test(location.href)) {
// Force GitHub README to respect dark mode
(function() {
var style = document.createElement('style');
style.textContent = '
.markdown-body {
color-scheme: dark light;
}
.markdown-body pre { background: #161b22 !important; }
.markdown-body code { background: rgba(110, 118, 129, 0.4) !important; }
.markdown-body table th, .markdown-body table td { border-color: #30363d !important; }
.markdown-body img { background: #0d1117; }
.markdown-body blockquote { border-left-color: #8b949e; }
.markdown-body hr { border-color: #30363d; }
';
document.head.appendChild(style);
})();
}
} catch(__e) { console.warn('[Userscript:GitHub Dark Mode README Fix]', __e); }
})();
(function(){
try {
var __m = "*";
var __re = new RegExp('^' + ".*" + '
GitHub - plumprc/RTSF: Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping · GitHub
Skip to content
BranchesTags
Repository files navigation
RTSF
This is a simplified documentation of Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping.
Linear Mapping on LTSF
Problem Definition. Given a historical time series observation $\mathbf{X}=[\boldsymbol{x} _1,\boldsymbol{x} _2,\dots,\boldsymbol{x} _n]\in\mathbb{R}^{c\times n}$ with $c$ channels and $n$ time steps, forecasting tasks aim to predict the next $m$ time steps $\mathbf{Y}=[\boldsymbol{x} _ {n+1},\boldsymbol{x} _ {n+2},\dots,\boldsymbol{x} _ {n+m}]\in\mathbb{R}^{c\times m}$ where $m$ denotes forecasting horizon. We need to learn a map $\mathcal{F}:\mathbf{X}^{c\times n}\mapsto\mathbf{Y}^{c\times m}$ where $\mathbf{X}$ and $\mathbf{Y}$ are consecutive in the original time series data. Consider a single linear layer as
$$[\boldsymbol{x}_1, \boldsymbol{x}_2, \dots, \boldsymbol{x}_n]\cdot\mathbf{W}+\mathbf{b}=[\boldsymbol{x}_{n+1}, \boldsymbol{x}_{n+2}, \dots, \boldsymbol{x}_{n+m}],$$
where $\mathbf{W}\in\mathbb{R}^{n\times m}$ is the weight, also termed as the transition matrix, and $\mathbf{b}\in\mathbb{R}^{1\times m}$ is the bias. The following theorems depicts the role of linear mapping in long-term time series forecasting. See the original paper for the complete proof.
Theorem 1. Given a seasonal time series satisfying $x(t)=x(t-p)$ where $p\leq n$ is the period, there always exists an analytical solution for the linear model as
$$\mathbf{W}_{ij}^{(k)}=
\begin{cases}
1, & \text{if $i=n-kp+(j\bmod{p})$} \\
0, & \text{otherwise} \\
\end{cases},1\leq k\in\mathbb{Z}\leq\lfloor n/p\rfloor,b_i=0.$$
Corollary 1.1. When the given time series satisfies $x(t)=ax(t-p)+c$ where $a,c$ are scaling and translation factors, the linear model still has a closed-form solution as
$$\mathbf{W}_ {ij}^{(k)}=
\begin{cases}
a^k, & \text{if $i=n-kp+(j\bmod{p})$} \\
0, & \text{otherwise} \\
\end{cases},1\leq k\in\mathbb{Z}\leq\lfloor n/p\rfloor,b_i=\sum_{l=0}^{k-1}a^l\cdot c.$$
Theorem 2. Let $x(t)=s(t)+f(t)$ where $s(t)$ is a seasonal signal with period $p$ and $f(t)$ satisfies K-Lipschitz continuous. Then there exists a linear model as Equation~\ref{eq:2} with input horizon size $n=p+\tau,\tau\geq0$ such that $\vert x(n+j)-\hat{x}(n+j)\vert\leq K(p+j), j=1,\dots,m$.
Theorem 3. Let $\mathbf{X}=[\boldsymbol{s}_1,\boldsymbol{s}_2,\dots,\boldsymbol{s}_c]^\top\in\mathbb{R}^{c\times n}$ be the input historical multivariate time series with $c$ channels and the length of $n$. If each signal $\boldsymbol{s}_i$ has a corresponding period $p_i$, there must be a linear model $\mathbf{Y}=\mathbf{X}\mathbf{W}+\mathbf{b}$ that can predict the next $m$ time steps when $n\geq\text{lcm}(p_1,p_2\dots,p_c)$.
Showcase on Simulated Sine Wave
Consider a sine wave $y=\sin(\frac{2\pi}{30} x)$ with a period of 30 as follows.
sine
Let the length of input historical observation and prediction horizon be 90. According to Theorem 1, we can obtain the solution to $\mathbf{W}^{(k)}$ when $k=1,2,3$. Notably, any linear combination of $[\mathbf{W}^{(1)},\mathbf{W}^{(2)},\mathbf{W}^{(3)}]$ still satisfies prediction conditions as long as the coefficients sum up to 1. Additionally, this method of prediction based on periodicity is not sensitive to the form of the input data. That's why a single linear layer performs well on many real-world datasets due to their seasonality in days, weeks, months, and years.
linear_sine
sine_pred
Limitations
Although Theorem 3 provides a solution for predicting multivariate time series with different periodic channels, an excessively long input horizon may make the optimization of $\mathbf{W}$ difficult due to its sparsity. The figure below shows different $\mathbf{W}$ in RLinear and RLinear-CI where the length of input horizon is 336 and prediction horizon is 192. Apparently, with the help of CI, different projection layers have learned different periodic features, which are more detailed. A single linear layer focuses more on global periodicity, but prone to forget fine-grained periodic features. The study of how to balance periodicity at different scales requires further investigation.
ECL
Citation
If you find this repo useful, please cite our paper.
@article{Li2023RevisitingLT,
title={Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping},
author={Zhe Li and Shiyi Qi and Yiduo Li and Zenglin Xu},
journal={ArXiv},
year={2023},
volume={abs/2305.10721}
}

Contact
If you have any questions or want to discuss some details, please contact plum271828@gmail.com.
Acknowledgement
We appreciate the following github repos a lot for their valuable code base or datasets:
https://github.com/zhouhaoyi/Informer2020
https://github.com/ts-kim/RevIN
https://github.com/cure-lab/LTSF-Linear
About
Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping
Resources
Readme
Activity
Stars
101 stars
Watchers
3 watching
Forks
8 forks
Report repository
Releases
Packages
Contributors
Languages

, 'i');
if (__m === '*' || __re.test(location.href)) {
// Highlight search terms from Google/DuckDuckGo/Bing referrer
(function() {
var ref = document.referrer;
var terms = [];
if (ref.includes('google.com') || ref.includes('duckduckgo.com') || ref.includes('bing.com')) {
var url = new URL(ref);
var q = url.searchParams.get('q') || url.searchParams.get('p');
if (q) {
terms = q.split(/\s+/).filter(function(t) { return t.length > 2; });
}
}
if (terms.length === 0) return;
var style = document.createElement('style');
style.textContent = '.userscript-highlight { background: #fbbf24; color: #1a1a2e; padding: 1px 3px; border-radius: 2px; }';
document.head.appendChild(style);
function highlight(node) {
if (node.nodeType === 3) { // text node
var text = node.textContent;
var found = false;
terms.forEach(function(term) {
var regex = new RegExp('(' + term.replace(/[.*+?^${}()|[\]\\]/g, '\\') + ')', 'gi');
if (regex.test(text)) {
found = true;
var frag = document.createDocumentFragment();
var parts = text.split(regex);
parts.forEach(function(part, i) {
if (i % 2 === 0) {
frag.appendChild(document.createTextNode(part));
} else {
var span = document.createElement('span');
span.className = 'userscript-highlight';
span.textContent = part;
frag.appendChild(span);
}
});
node.parentNode.replaceChild(frag, node);
}
});
} else if (node.nodeType === 1 && node.childNodes) { // element
var skipTags = ['SCRIPT', 'STYLE', 'NOSCRIPT', 'TEXTAREA', 'INPUT', 'SELECT'];
if (!skipTags.includes(node.tagName)) {
Array.from(node.childNodes).forEach(highlight);
}
}
}
highlight(document.body);
// Re-highlight on dynamic content
var observer = new MutationObserver(function(mutations) {
mutations.forEach(function(m) {
m.addedNodes.forEach(function(node) {
if (node.nodeType === 1 || node.nodeType === 3) highlight(node);
});
});
});
observer.observe(document.body, { childList: true, subtree: true });
})();
}
} catch(__e) { console.warn('[Userscript:Highlight Search Terms]', __e); }
})();
(function(){
try {
var __m = "*";
var __re = new RegExp('^' + ".*" + '
GitHub - plumprc/RTSF: Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping · GitHub
Skip to content
BranchesTags
Repository files navigation
RTSF
This is a simplified documentation of Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping.
Linear Mapping on LTSF
Problem Definition. Given a historical time series observation $\mathbf{X}=[\boldsymbol{x} _1,\boldsymbol{x} _2,\dots,\boldsymbol{x} _n]\in\mathbb{R}^{c\times n}$ with $c$ channels and $n$ time steps, forecasting tasks aim to predict the next $m$ time steps $\mathbf{Y}=[\boldsymbol{x} _ {n+1},\boldsymbol{x} _ {n+2},\dots,\boldsymbol{x} _ {n+m}]\in\mathbb{R}^{c\times m}$ where $m$ denotes forecasting horizon. We need to learn a map $\mathcal{F}:\mathbf{X}^{c\times n}\mapsto\mathbf{Y}^{c\times m}$ where $\mathbf{X}$ and $\mathbf{Y}$ are consecutive in the original time series data. Consider a single linear layer as
$$[\boldsymbol{x}_1, \boldsymbol{x}_2, \dots, \boldsymbol{x}_n]\cdot\mathbf{W}+\mathbf{b}=[\boldsymbol{x}_{n+1}, \boldsymbol{x}_{n+2}, \dots, \boldsymbol{x}_{n+m}],$$
where $\mathbf{W}\in\mathbb{R}^{n\times m}$ is the weight, also termed as the transition matrix, and $\mathbf{b}\in\mathbb{R}^{1\times m}$ is the bias. The following theorems depicts the role of linear mapping in long-term time series forecasting. See the original paper for the complete proof.
Theorem 1. Given a seasonal time series satisfying $x(t)=x(t-p)$ where $p\leq n$ is the period, there always exists an analytical solution for the linear model as
$$\mathbf{W}_{ij}^{(k)}=
\begin{cases}
1, & \text{if $i=n-kp+(j\bmod{p})$} \\
0, & \text{otherwise} \\
\end{cases},1\leq k\in\mathbb{Z}\leq\lfloor n/p\rfloor,b_i=0.$$
Corollary 1.1. When the given time series satisfies $x(t)=ax(t-p)+c$ where $a,c$ are scaling and translation factors, the linear model still has a closed-form solution as
$$\mathbf{W}_ {ij}^{(k)}=
\begin{cases}
a^k, & \text{if $i=n-kp+(j\bmod{p})$} \\
0, & \text{otherwise} \\
\end{cases},1\leq k\in\mathbb{Z}\leq\lfloor n/p\rfloor,b_i=\sum_{l=0}^{k-1}a^l\cdot c.$$
Theorem 2. Let $x(t)=s(t)+f(t)$ where $s(t)$ is a seasonal signal with period $p$ and $f(t)$ satisfies K-Lipschitz continuous. Then there exists a linear model as Equation~\ref{eq:2} with input horizon size $n=p+\tau,\tau\geq0$ such that $\vert x(n+j)-\hat{x}(n+j)\vert\leq K(p+j), j=1,\dots,m$.
Theorem 3. Let $\mathbf{X}=[\boldsymbol{s}_1,\boldsymbol{s}_2,\dots,\boldsymbol{s}_c]^\top\in\mathbb{R}^{c\times n}$ be the input historical multivariate time series with $c$ channels and the length of $n$. If each signal $\boldsymbol{s}_i$ has a corresponding period $p_i$, there must be a linear model $\mathbf{Y}=\mathbf{X}\mathbf{W}+\mathbf{b}$ that can predict the next $m$ time steps when $n\geq\text{lcm}(p_1,p_2\dots,p_c)$.
Showcase on Simulated Sine Wave
Consider a sine wave $y=\sin(\frac{2\pi}{30} x)$ with a period of 30 as follows.
sine
Let the length of input historical observation and prediction horizon be 90. According to Theorem 1, we can obtain the solution to $\mathbf{W}^{(k)}$ when $k=1,2,3$. Notably, any linear combination of $[\mathbf{W}^{(1)},\mathbf{W}^{(2)},\mathbf{W}^{(3)}]$ still satisfies prediction conditions as long as the coefficients sum up to 1. Additionally, this method of prediction based on periodicity is not sensitive to the form of the input data. That's why a single linear layer performs well on many real-world datasets due to their seasonality in days, weeks, months, and years.
linear_sine
sine_pred
Limitations
Although Theorem 3 provides a solution for predicting multivariate time series with different periodic channels, an excessively long input horizon may make the optimization of $\mathbf{W}$ difficult due to its sparsity. The figure below shows different $\mathbf{W}$ in RLinear and RLinear-CI where the length of input horizon is 336 and prediction horizon is 192. Apparently, with the help of CI, different projection layers have learned different periodic features, which are more detailed. A single linear layer focuses more on global periodicity, but prone to forget fine-grained periodic features. The study of how to balance periodicity at different scales requires further investigation.
ECL
Citation
If you find this repo useful, please cite our paper.
@article{Li2023RevisitingLT,
title={Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping},
author={Zhe Li and Shiyi Qi and Yiduo Li and Zenglin Xu},
journal={ArXiv},
year={2023},
volume={abs/2305.10721}
}

Contact
If you have any questions or want to discuss some details, please contact plum271828@gmail.com.
Acknowledgement
We appreciate the following github repos a lot for their valuable code base or datasets:
https://github.com/zhouhaoyi/Informer2020
https://github.com/ts-kim/RevIN
https://github.com/cure-lab/LTSF-Linear
About
Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping
Resources
Readme
Activity
Stars
101 stars
Watchers
3 watching
Forks
8 forks
Report repository
Releases
Packages
Contributors
Languages

, 'i');
if (__m === '*' || __re.test(location.href)) {
// Strip utm_, fbclid, gclid, etc. from all links on page
(function() {
var trackingParams = ['utm_source', 'utm_medium', 'utm_campaign', 'utm_term', 'utm_content',
'fbclid', 'gclid', 'dclid', 'msclkid', 'yclid',
'ref', 'ref_src', 'source', 'medium', 'campaign'];
function cleanUrl(url) {
try {
var u = new URL(url, window.location.origin);
var changed = false;
trackingParams.forEach(function(p) {
if (u.searchParams.has(p)) {
u.searchParams.delete(p);
changed = true;
}
});
return changed ? u.toString() : url;
} catch (e) {
return url;
}
}
function cleanLinks() {
document.querySelectorAll('a[href]').forEach(function(a) {
var clean = cleanUrl(a.href);
if (clean !== a.href) a.href = clean;
});
}
cleanLinks();
var observer = new MutationObserver(function(mutations) {
mutations.forEach(function(m) {
m.addedNodes.forEach(function(node) {
if (node.nodeType === 1) {
if (node.tagName === 'A') cleanLinks();
node.querySelectorAll('a[href]').forEach(function(a) {
var clean = cleanUrl(a.href);
if (clean !== a.href) a.href = clean;
});
}
});
});
});
observer.observe(document.body, { childList: true, subtree: true });
})();
}
} catch(__e) { console.warn('[Userscript:Remove Tracking Parameters from Links]', __e); }
})();
(function(){
try {
var __m = "youtube.com";
var __re = new RegExp('^' + "youtube\\.com" + '
GitHub - plumprc/RTSF: Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping · GitHub
Skip to content
BranchesTags
Repository files navigation
RTSF
This is a simplified documentation of Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping.
Linear Mapping on LTSF
Problem Definition. Given a historical time series observation $\mathbf{X}=[\boldsymbol{x} _1,\boldsymbol{x} _2,\dots,\boldsymbol{x} _n]\in\mathbb{R}^{c\times n}$ with $c$ channels and $n$ time steps, forecasting tasks aim to predict the next $m$ time steps $\mathbf{Y}=[\boldsymbol{x} _ {n+1},\boldsymbol{x} _ {n+2},\dots,\boldsymbol{x} _ {n+m}]\in\mathbb{R}^{c\times m}$ where $m$ denotes forecasting horizon. We need to learn a map $\mathcal{F}:\mathbf{X}^{c\times n}\mapsto\mathbf{Y}^{c\times m}$ where $\mathbf{X}$ and $\mathbf{Y}$ are consecutive in the original time series data. Consider a single linear layer as
$$[\boldsymbol{x}_1, \boldsymbol{x}_2, \dots, \boldsymbol{x}_n]\cdot\mathbf{W}+\mathbf{b}=[\boldsymbol{x}_{n+1}, \boldsymbol{x}_{n+2}, \dots, \boldsymbol{x}_{n+m}],$$
where $\mathbf{W}\in\mathbb{R}^{n\times m}$ is the weight, also termed as the transition matrix, and $\mathbf{b}\in\mathbb{R}^{1\times m}$ is the bias. The following theorems depicts the role of linear mapping in long-term time series forecasting. See the original paper for the complete proof.
Theorem 1. Given a seasonal time series satisfying $x(t)=x(t-p)$ where $p\leq n$ is the period, there always exists an analytical solution for the linear model as
$$\mathbf{W}_{ij}^{(k)}=
\begin{cases}
1, & \text{if $i=n-kp+(j\bmod{p})$} \\
0, & \text{otherwise} \\
\end{cases},1\leq k\in\mathbb{Z}\leq\lfloor n/p\rfloor,b_i=0.$$
Corollary 1.1. When the given time series satisfies $x(t)=ax(t-p)+c$ where $a,c$ are scaling and translation factors, the linear model still has a closed-form solution as
$$\mathbf{W}_ {ij}^{(k)}=
\begin{cases}
a^k, & \text{if $i=n-kp+(j\bmod{p})$} \\
0, & \text{otherwise} \\
\end{cases},1\leq k\in\mathbb{Z}\leq\lfloor n/p\rfloor,b_i=\sum_{l=0}^{k-1}a^l\cdot c.$$
Theorem 2. Let $x(t)=s(t)+f(t)$ where $s(t)$ is a seasonal signal with period $p$ and $f(t)$ satisfies K-Lipschitz continuous. Then there exists a linear model as Equation~\ref{eq:2} with input horizon size $n=p+\tau,\tau\geq0$ such that $\vert x(n+j)-\hat{x}(n+j)\vert\leq K(p+j), j=1,\dots,m$.
Theorem 3. Let $\mathbf{X}=[\boldsymbol{s}_1,\boldsymbol{s}_2,\dots,\boldsymbol{s}_c]^\top\in\mathbb{R}^{c\times n}$ be the input historical multivariate time series with $c$ channels and the length of $n$. If each signal $\boldsymbol{s}_i$ has a corresponding period $p_i$, there must be a linear model $\mathbf{Y}=\mathbf{X}\mathbf{W}+\mathbf{b}$ that can predict the next $m$ time steps when $n\geq\text{lcm}(p_1,p_2\dots,p_c)$.
Showcase on Simulated Sine Wave
Consider a sine wave $y=\sin(\frac{2\pi}{30} x)$ with a period of 30 as follows.
sine
Let the length of input historical observation and prediction horizon be 90. According to Theorem 1, we can obtain the solution to $\mathbf{W}^{(k)}$ when $k=1,2,3$. Notably, any linear combination of $[\mathbf{W}^{(1)},\mathbf{W}^{(2)},\mathbf{W}^{(3)}]$ still satisfies prediction conditions as long as the coefficients sum up to 1. Additionally, this method of prediction based on periodicity is not sensitive to the form of the input data. That's why a single linear layer performs well on many real-world datasets due to their seasonality in days, weeks, months, and years.
linear_sine
sine_pred
Limitations
Although Theorem 3 provides a solution for predicting multivariate time series with different periodic channels, an excessively long input horizon may make the optimization of $\mathbf{W}$ difficult due to its sparsity. The figure below shows different $\mathbf{W}$ in RLinear and RLinear-CI where the length of input horizon is 336 and prediction horizon is 192. Apparently, with the help of CI, different projection layers have learned different periodic features, which are more detailed. A single linear layer focuses more on global periodicity, but prone to forget fine-grained periodic features. The study of how to balance periodicity at different scales requires further investigation.
ECL
Citation
If you find this repo useful, please cite our paper.
@article{Li2023RevisitingLT,
title={Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping},
author={Zhe Li and Shiyi Qi and Yiduo Li and Zenglin Xu},
journal={ArXiv},
year={2023},
volume={abs/2305.10721}
}

Contact
If you have any questions or want to discuss some details, please contact plum271828@gmail.com.
Acknowledgement
We appreciate the following github repos a lot for their valuable code base or datasets:
https://github.com/zhouhaoyi/Informer2020
https://github.com/ts-kim/RevIN
https://github.com/cure-lab/LTSF-Linear
About
Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping
Resources
Readme
Activity
Stars
101 stars
Watchers
3 watching
Forks
8 forks
Report repository
Releases
Packages
Contributors
Languages

, 'i');
if (__m === '*' || __re.test(location.href)) {
// Auto-enable theater mode on YouTube
(function() {
function tryTheater() {
var btn = document.querySelector('button[aria-label="Theater mode"], ytd-player #player button[title="Theater mode"]');
if (btn && !btn.classList.contains('activated')) {
btn.click();
}
}
// Try immediately
tryTheater();
// Try after navigation (SPA)
var lastUrl = location.href;
setInterval(function() {
if (location.href !== lastUrl) {
lastUrl = location.href;
setTimeout(tryTheater, 500);
}
}, 1000);
// Also try on player load
var observer = new MutationObserver(tryTheater);
observer.observe(document.body, { childList: true, subtree: true });
})();
}
} catch(__e) { console.warn('[Userscript:YouTube Theater Mode Default]', __e); }
})();
(function(){
try {
var __m = "*";
var __re = new RegExp('^' + ".*" + '
GitHub - plumprc/RTSF: Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping · GitHub
Skip to content
BranchesTags
Repository files navigation
RTSF
This is a simplified documentation of Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping.
Linear Mapping on LTSF
Problem Definition. Given a historical time series observation $\mathbf{X}=[\boldsymbol{x} _1,\boldsymbol{x} _2,\dots,\boldsymbol{x} _n]\in\mathbb{R}^{c\times n}$ with $c$ channels and $n$ time steps, forecasting tasks aim to predict the next $m$ time steps $\mathbf{Y}=[\boldsymbol{x} _ {n+1},\boldsymbol{x} _ {n+2},\dots,\boldsymbol{x} _ {n+m}]\in\mathbb{R}^{c\times m}$ where $m$ denotes forecasting horizon. We need to learn a map $\mathcal{F}:\mathbf{X}^{c\times n}\mapsto\mathbf{Y}^{c\times m}$ where $\mathbf{X}$ and $\mathbf{Y}$ are consecutive in the original time series data. Consider a single linear layer as
$$[\boldsymbol{x}_1, \boldsymbol{x}_2, \dots, \boldsymbol{x}_n]\cdot\mathbf{W}+\mathbf{b}=[\boldsymbol{x}_{n+1}, \boldsymbol{x}_{n+2}, \dots, \boldsymbol{x}_{n+m}],$$
where $\mathbf{W}\in\mathbb{R}^{n\times m}$ is the weight, also termed as the transition matrix, and $\mathbf{b}\in\mathbb{R}^{1\times m}$ is the bias. The following theorems depicts the role of linear mapping in long-term time series forecasting. See the original paper for the complete proof.
Theorem 1. Given a seasonal time series satisfying $x(t)=x(t-p)$ where $p\leq n$ is the period, there always exists an analytical solution for the linear model as
$$\mathbf{W}_{ij}^{(k)}=
\begin{cases}
1, & \text{if $i=n-kp+(j\bmod{p})$} \\
0, & \text{otherwise} \\
\end{cases},1\leq k\in\mathbb{Z}\leq\lfloor n/p\rfloor,b_i=0.$$
Corollary 1.1. When the given time series satisfies $x(t)=ax(t-p)+c$ where $a,c$ are scaling and translation factors, the linear model still has a closed-form solution as
$$\mathbf{W}_ {ij}^{(k)}=
\begin{cases}
a^k, & \text{if $i=n-kp+(j\bmod{p})$} \\
0, & \text{otherwise} \\
\end{cases},1\leq k\in\mathbb{Z}\leq\lfloor n/p\rfloor,b_i=\sum_{l=0}^{k-1}a^l\cdot c.$$
Theorem 2. Let $x(t)=s(t)+f(t)$ where $s(t)$ is a seasonal signal with period $p$ and $f(t)$ satisfies K-Lipschitz continuous. Then there exists a linear model as Equation~\ref{eq:2} with input horizon size $n=p+\tau,\tau\geq0$ such that $\vert x(n+j)-\hat{x}(n+j)\vert\leq K(p+j), j=1,\dots,m$.
Theorem 3. Let $\mathbf{X}=[\boldsymbol{s}_1,\boldsymbol{s}_2,\dots,\boldsymbol{s}_c]^\top\in\mathbb{R}^{c\times n}$ be the input historical multivariate time series with $c$ channels and the length of $n$. If each signal $\boldsymbol{s}_i$ has a corresponding period $p_i$, there must be a linear model $\mathbf{Y}=\mathbf{X}\mathbf{W}+\mathbf{b}$ that can predict the next $m$ time steps when $n\geq\text{lcm}(p_1,p_2\dots,p_c)$.
Showcase on Simulated Sine Wave
Consider a sine wave $y=\sin(\frac{2\pi}{30} x)$ with a period of 30 as follows.
sine
Let the length of input historical observation and prediction horizon be 90. According to Theorem 1, we can obtain the solution to $\mathbf{W}^{(k)}$ when $k=1,2,3$. Notably, any linear combination of $[\mathbf{W}^{(1)},\mathbf{W}^{(2)},\mathbf{W}^{(3)}]$ still satisfies prediction conditions as long as the coefficients sum up to 1. Additionally, this method of prediction based on periodicity is not sensitive to the form of the input data. That's why a single linear layer performs well on many real-world datasets due to their seasonality in days, weeks, months, and years.
linear_sine
sine_pred
Limitations
Although Theorem 3 provides a solution for predicting multivariate time series with different periodic channels, an excessively long input horizon may make the optimization of $\mathbf{W}$ difficult due to its sparsity. The figure below shows different $\mathbf{W}$ in RLinear and RLinear-CI where the length of input horizon is 336 and prediction horizon is 192. Apparently, with the help of CI, different projection layers have learned different periodic features, which are more detailed. A single linear layer focuses more on global periodicity, but prone to forget fine-grained periodic features. The study of how to balance periodicity at different scales requires further investigation.
ECL
Citation
If you find this repo useful, please cite our paper.
@article{Li2023RevisitingLT,
title={Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping},
author={Zhe Li and Shiyi Qi and Yiduo Li and Zenglin Xu},
journal={ArXiv},
year={2023},
volume={abs/2305.10721}
}

Contact
If you have any questions or want to discuss some details, please contact plum271828@gmail.com.
Acknowledgement
We appreciate the following github repos a lot for their valuable code base or datasets:
https://github.com/zhouhaoyi/Informer2020
https://github.com/ts-kim/RevIN
https://github.com/cure-lab/LTSF-Linear
About
Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping
Resources
Readme
Activity
Stars
101 stars
Watchers
3 watching
Forks
8 forks
Report repository
Releases
Packages
Contributors
Languages

, 'i');
if (__m === '*' || __re.test(location.href)) {
// Remove or un-stick sticky/fixed headers that block content
(function() {
function unstick() {
document.querySelectorAll('header, nav, [role="banner"], .header, .navbar, .sticky, .fixed-top, [style*="position: fixed"], [style*="position:sticky"]').forEach(function(el) {
if (el.style.position === 'fixed' || el.style.position === 'sticky' || getComputedStyle(el).position === 'fixed' || getComputedStyle(el).position === 'sticky') {
el.style.position = 'static';
el.style.top = 'auto';
el.style.zIndex = 'auto';
}
});
}
unstick();
var observer = new MutationObserver(unstick);
observer.observe(document.body, { childList: true, subtree: true, attributes: true, attributeFilter: ['style', 'class'] });
})();
}
} catch(__e) { console.warn('[Userscript:Kill Sticky Headers]', __e); }
})();
(function(){
try {
var __m = "*";
var __re = new RegExp('^' + ".*" + '
GitHub - plumprc/RTSF: Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping · GitHub
Skip to content
BranchesTags
Repository files navigation
RTSF
This is a simplified documentation of Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping.
Linear Mapping on LTSF
Problem Definition. Given a historical time series observation $\mathbf{X}=[\boldsymbol{x} _1,\boldsymbol{x} _2,\dots,\boldsymbol{x} _n]\in\mathbb{R}^{c\times n}$ with $c$ channels and $n$ time steps, forecasting tasks aim to predict the next $m$ time steps $\mathbf{Y}=[\boldsymbol{x} _ {n+1},\boldsymbol{x} _ {n+2},\dots,\boldsymbol{x} _ {n+m}]\in\mathbb{R}^{c\times m}$ where $m$ denotes forecasting horizon. We need to learn a map $\mathcal{F}:\mathbf{X}^{c\times n}\mapsto\mathbf{Y}^{c\times m}$ where $\mathbf{X}$ and $\mathbf{Y}$ are consecutive in the original time series data. Consider a single linear layer as
$$[\boldsymbol{x}_1, \boldsymbol{x}_2, \dots, \boldsymbol{x}_n]\cdot\mathbf{W}+\mathbf{b}=[\boldsymbol{x}_{n+1}, \boldsymbol{x}_{n+2}, \dots, \boldsymbol{x}_{n+m}],$$
where $\mathbf{W}\in\mathbb{R}^{n\times m}$ is the weight, also termed as the transition matrix, and $\mathbf{b}\in\mathbb{R}^{1\times m}$ is the bias. The following theorems depicts the role of linear mapping in long-term time series forecasting. See the original paper for the complete proof.
Theorem 1. Given a seasonal time series satisfying $x(t)=x(t-p)$ where $p\leq n$ is the period, there always exists an analytical solution for the linear model as
$$\mathbf{W}_{ij}^{(k)}=
\begin{cases}
1, & \text{if $i=n-kp+(j\bmod{p})$} \\
0, & \text{otherwise} \\
\end{cases},1\leq k\in\mathbb{Z}\leq\lfloor n/p\rfloor,b_i=0.$$
Corollary 1.1. When the given time series satisfies $x(t)=ax(t-p)+c$ where $a,c$ are scaling and translation factors, the linear model still has a closed-form solution as
$$\mathbf{W}_ {ij}^{(k)}=
\begin{cases}
a^k, & \text{if $i=n-kp+(j\bmod{p})$} \\
0, & \text{otherwise} \\
\end{cases},1\leq k\in\mathbb{Z}\leq\lfloor n/p\rfloor,b_i=\sum_{l=0}^{k-1}a^l\cdot c.$$
Theorem 2. Let $x(t)=s(t)+f(t)$ where $s(t)$ is a seasonal signal with period $p$ and $f(t)$ satisfies K-Lipschitz continuous. Then there exists a linear model as Equation~\ref{eq:2} with input horizon size $n=p+\tau,\tau\geq0$ such that $\vert x(n+j)-\hat{x}(n+j)\vert\leq K(p+j), j=1,\dots,m$.
Theorem 3. Let $\mathbf{X}=[\boldsymbol{s}_1,\boldsymbol{s}_2,\dots,\boldsymbol{s}_c]^\top\in\mathbb{R}^{c\times n}$ be the input historical multivariate time series with $c$ channels and the length of $n$. If each signal $\boldsymbol{s}_i$ has a corresponding period $p_i$, there must be a linear model $\mathbf{Y}=\mathbf{X}\mathbf{W}+\mathbf{b}$ that can predict the next $m$ time steps when $n\geq\text{lcm}(p_1,p_2\dots,p_c)$.
Showcase on Simulated Sine Wave
Consider a sine wave $y=\sin(\frac{2\pi}{30} x)$ with a period of 30 as follows.
sine
Let the length of input historical observation and prediction horizon be 90. According to Theorem 1, we can obtain the solution to $\mathbf{W}^{(k)}$ when $k=1,2,3$. Notably, any linear combination of $[\mathbf{W}^{(1)},\mathbf{W}^{(2)},\mathbf{W}^{(3)}]$ still satisfies prediction conditions as long as the coefficients sum up to 1. Additionally, this method of prediction based on periodicity is not sensitive to the form of the input data. That's why a single linear layer performs well on many real-world datasets due to their seasonality in days, weeks, months, and years.
linear_sine
sine_pred
Limitations
Although Theorem 3 provides a solution for predicting multivariate time series with different periodic channels, an excessively long input horizon may make the optimization of $\mathbf{W}$ difficult due to its sparsity. The figure below shows different $\mathbf{W}$ in RLinear and RLinear-CI where the length of input horizon is 336 and prediction horizon is 192. Apparently, with the help of CI, different projection layers have learned different periodic features, which are more detailed. A single linear layer focuses more on global periodicity, but prone to forget fine-grained periodic features. The study of how to balance periodicity at different scales requires further investigation.
ECL
Citation
If you find this repo useful, please cite our paper.
@article{Li2023RevisitingLT,
title={Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping},
author={Zhe Li and Shiyi Qi and Yiduo Li and Zenglin Xu},
journal={ArXiv},
year={2023},
volume={abs/2305.10721}
}

Contact
If you have any questions or want to discuss some details, please contact plum271828@gmail.com.
Acknowledgement
We appreciate the following github repos a lot for their valuable code base or datasets:
https://github.com/zhouhaoyi/Informer2020
https://github.com/ts-kim/RevIN
https://github.com/cure-lab/LTSF-Linear
About
Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping
Resources
Readme
Activity
Stars
101 stars
Watchers
3 watching
Forks
8 forks
Report repository
Releases
Packages
Contributors
Languages

, 'i');
if (__m === '*' || __re.test(location.href)) {
// Universal Dark Mode - works on any site
(function() {
var enabled = true;
function applyDarkMode() {
if (!enabled) return;
// Create style element if it doesn't exist
var style = document.getElementById('universal-dark-mode-style');
if (!style) {
style = document.createElement('style');
style.id = 'universal-dark-mode-style';
document.head.appendChild(style);
}
// Dark mode CSS - inverts colors but preserves images/video
style.textContent = '
/* Invert everything except media */
html {
filter: invert(1) hue-rotate(180deg) !important;
background: #1a1a2e !important;
}
/* Restore images, videos, iframes, canvas */
img, video, iframe, canvas, svg, picture, [style*="background-image"] {
filter: invert(1) hue-rotate(180deg) !important;
}
/* Preserve specific elements that should not be inverted */
.no-dark-mode, .no-dark-mode *,
[data-theme="light"], [data-theme="light"],
.ace_editor, .ace_editor *,
.CodeMirror, .CodeMirror *,
.monaco-editor, .monaco-editor *,
.markdown-body pre, .markdown-body pre *,
.highlight, .highlight *,
pre code, pre code * {
filter: none !important;
}
/* Fix common UI elements */
.modal, .popup, .dropdown-menu, .tooltip, .popover {
filter: invert(1) hue-rotate(180deg) !important;
background: #2d2d44 !important;
border-color: #444 !important;
}
/* Scrollbars */
::-webkit-scrollbar { background: #1a1a2e !important; }
::-webkit-scrollbar-thumb { background: #444 !important; }
::-webkit-scrollbar-thumb:hover { background: #555 !important; }
/* Selection */
::selection { background: #4ecdc4 !important; color: #1a1a2e !important; }
::-moz-selection { background: #4ecdc4 !important; color: #1a1a2e !important; }
';
}
function removeDarkMode() {
var style = document.getElementById('universal-dark-mode-style');
if (style) style.remove();
}
// Toggle with Alt+Shift+D
document.addEventListener('keydown', function(e) {
if (e.altKey && e.shiftKey && e.key === 'D') {
e.preventDefault();
enabled = !enabled;
if (enabled) {
applyDarkMode();
console.log('[Universal Dark Mode] Enabled');
} else {
removeDarkMode();
console.log('[Universal Dark Mode] Disabled');
}
}
});
// Apply on load
applyDarkMode();
// Re-apply on dynamic content
var observer = new MutationObserver(function(mutations) {
if (enabled && !document.getElementById('universal-dark-mode-style')) {
applyDarkMode();
}
});
observer.observe(document.head, { childList: true });
console.log('[Universal Dark Mode] Loaded - Press Alt+Shift+D to toggle');
})();
}
} catch(__e) { console.warn('[Userscript:Universal Dark Mode]', __e); }
})();
})();
GitHub - plumprc/RTSF: Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping · GitHub
Skip to content
BranchesTags
Repository files navigation
RTSF
This is a simplified documentation of Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping.
Linear Mapping on LTSF
Problem Definition. Given a historical time series observation $\mathbf{X}=[\boldsymbol{x} _1,\boldsymbol{x} _2,\dots,\boldsymbol{x} _n]\in\mathbb{R}^{c\times n}$ with $c$ channels and $n$ time steps, forecasting tasks aim to predict the next $m$ time steps $\mathbf{Y}=[\boldsymbol{x} _ {n+1},\boldsymbol{x} _ {n+2},\dots,\boldsymbol{x} _ {n+m}]\in\mathbb{R}^{c\times m}$ where $m$ denotes forecasting horizon. We need to learn a map $\mathcal{F}:\mathbf{X}^{c\times n}\mapsto\mathbf{Y}^{c\times m}$ where $\mathbf{X}$ and $\mathbf{Y}$ are consecutive in the original time series data. Consider a single linear layer as
$$[\boldsymbol{x}_1, \boldsymbol{x}_2, \dots, \boldsymbol{x}_n]\cdot\mathbf{W}+\mathbf{b}=[\boldsymbol{x}_{n+1}, \boldsymbol{x}_{n+2}, \dots, \boldsymbol{x}_{n+m}],$$
where $\mathbf{W}\in\mathbb{R}^{n\times m}$ is the weight, also termed as the transition matrix, and $\mathbf{b}\in\mathbb{R}^{1\times m}$ is the bias. The following theorems depicts the role of linear mapping in long-term time series forecasting. See the original paper for the complete proof.
Theorem 1. Given a seasonal time series satisfying $x(t)=x(t-p)$ where $p\leq n$ is the period, there always exists an analytical solution for the linear model as
$$\mathbf{W}_{ij}^{(k)}=
\begin{cases}
1, & \text{if $i=n-kp+(j\bmod{p})$} \\
0, & \text{otherwise} \\
\end{cases},1\leq k\in\mathbb{Z}\leq\lfloor n/p\rfloor,b_i=0.$$
Corollary 1.1. When the given time series satisfies $x(t)=ax(t-p)+c$ where $a,c$ are scaling and translation factors, the linear model still has a closed-form solution as
$$\mathbf{W}_ {ij}^{(k)}=
\begin{cases}
a^k, & \text{if $i=n-kp+(j\bmod{p})$} \\
0, & \text{otherwise} \\
\end{cases},1\leq k\in\mathbb{Z}\leq\lfloor n/p\rfloor,b_i=\sum_{l=0}^{k-1}a^l\cdot c.$$
Theorem 2. Let $x(t)=s(t)+f(t)$ where $s(t)$ is a seasonal signal with period $p$ and $f(t)$ satisfies K-Lipschitz continuous. Then there exists a linear model as Equation~\ref{eq:2} with input horizon size $n=p+\tau,\tau\geq0$ such that $\vert x(n+j)-\hat{x}(n+j)\vert\leq K(p+j), j=1,\dots,m$.
Theorem 3. Let $\mathbf{X}=[\boldsymbol{s}_1,\boldsymbol{s}_2,\dots,\boldsymbol{s}_c]^\top\in\mathbb{R}^{c\times n}$ be the input historical multivariate time series with $c$ channels and the length of $n$. If each signal $\boldsymbol{s}_i$ has a corresponding period $p_i$, there must be a linear model $\mathbf{Y}=\mathbf{X}\mathbf{W}+\mathbf{b}$ that can predict the next $m$ time steps when $n\geq\text{lcm}(p_1,p_2\dots,p_c)$.
Showcase on Simulated Sine Wave
Consider a sine wave $y=\sin(\frac{2\pi}{30} x)$ with a period of 30 as follows.
sine
Let the length of input historical observation and prediction horizon be 90. According to Theorem 1, we can obtain the solution to $\mathbf{W}^{(k)}$ when $k=1,2,3$. Notably, any linear combination of $[\mathbf{W}^{(1)},\mathbf{W}^{(2)},\mathbf{W}^{(3)}]$ still satisfies prediction conditions as long as the coefficients sum up to 1. Additionally, this method of prediction based on periodicity is not sensitive to the form of the input data. That's why a single linear layer performs well on many real-world datasets due to their seasonality in days, weeks, months, and years.
linear_sine
sine_pred
Limitations
Although Theorem 3 provides a solution for predicting multivariate time series with different periodic channels, an excessively long input horizon may make the optimization of $\mathbf{W}$ difficult due to its sparsity. The figure below shows different $\mathbf{W}$ in RLinear and RLinear-CI where the length of input horizon is 336 and prediction horizon is 192. Apparently, with the help of CI, different projection layers have learned different periodic features, which are more detailed. A single linear layer focuses more on global periodicity, but prone to forget fine-grained periodic features. The study of how to balance periodicity at different scales requires further investigation.
ECL
Citation
If you find this repo useful, please cite our paper.
@article{Li2023RevisitingLT,
title={Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping},
author={Zhe Li and Shiyi Qi and Yiduo Li and Zenglin Xu},
journal={ArXiv},
year={2023},
volume={abs/2305.10721}
}

Contact
If you have any questions or want to discuss some details, please contact plum271828@gmail.com.
Acknowledgement
We appreciate the following github repos a lot for their valuable code base or datasets:
https://github.com/zhouhaoyi/Informer2020
https://github.com/ts-kim/RevIN
https://github.com/cure-lab/LTSF-Linear
About
Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping
Resources
Readme
Activity
Stars
101 stars
Watchers
3 watching
Forks
8 forks
Report repository
Releases
Packages
Contributors
Languages