# Approximate ar function so that it becomes linear in parameters (without Taylor)

TL;DR (original question):

I am looking for a function that has roughly the form of

$$f(x) = \exp\left(−(x/a)^2\right) − \log\left((x/a)^2+1\right)$$

but is linear in its model parameter (here $$a$$) so that I can use linear least squares to determine the parameter. Is there such a function?

EDIT (more explanations):

It seems there are a couple of misunderstandings:

# Background

I do not ask for for "linear regression" (fitting a line through points). Instead I want to do a non-linear regression with the function $$f$$ (fitting $$f$$ through points). In order to do a regression, one needs to perform a mathematical optimization - either with a numerical iterative method (e.g. Newton's method) or (if available) with an analytical closed form formula.

Most of the time, regression can be done by formulating a least-squares optimization cost function

$$J(\Theta) = \sum_i \left( y_i - f(x_i | \Theta) \right)^2$$ where \Theta is the collection of all optimization variables. Here I only have $$a$$. The optimization solves the problem $$\min_{\Theta} J(\Theta)$$

If this cost function is linear in the optimization variables, it can be solved with a closed form formula. If not, afaik, numerical iterative optimization is required. That is what I want to avoid due to the low speed.

For that I see two options of approximation:

1. reformulate the problem
2. reformulate $$f$$

## reformulate the problem

Let's see an example for the first case and choose Logistic Regression, which has the original (not approximated) problem formulation

\begin{align} f(x) &= \frac{1}{1 + exp(-\Theta\cdot x)} \\ J(\Theta) &= \sum_i \left( y_i - f(x_i | \Theta) \right)^2 = \sum_i \left( y_i - \frac{1}{1 + exp(-\Theta\cdot x_i)} \right)^2\\ \min_{\Theta} & \Big( J(\Theta) \Big) \end{align}

This can only be solved using iterative optimization, but it can be solved in closed form after reformulating it to

\begin{align} log\left( \frac{1}{f(x)} - 1 \right) &= -\Theta\cdot x \\ J(\Theta) &= \sum_i \left( log\left( \frac{1}{y_i} - 1 \right) + \Theta\cdot _i \right)^2\\ \min_{\Theta} & \Big( J(\Theta) \Big) \end{align}

So in other words we have the linear equation system, where one of the rows of this system is

$$log\left( \frac{1}{y_i} - 1 \right) = \Theta\cdot x_i$$

This system can be solved in closed form and which approximately solves the original problem. (I am not being super precise here, but I hope it is good enough.)

## reformulate $$f$$

An obvious solution is to approximate $$f$$ using Taylor, but I am hoping for something better. One reason is that Taylor for multivariate functions is a real pain.

Thus I am asking for an alternative function $$g(x)\approx f(x)$$ which is linear in the function parameters. An If there is a way to write $$f$$ such that linear least-squares can be performed (and thus the closed form)

# Back to the question and why I asked it.

Honestly, I have my doubt that the approach I called "reformulate the problem" will be possible here, but this would be preferred. If someone has an idea, I would be very happy.

That is why I was asking for an approach of the type "reformulate $$f$$".

The answer that Claude Leibovici provides is neither of those approaches exactly and it does not eliminate the need for iterative optimization. That being said, it is kind of close to the approach of "reformulate the problem", since it does indeed reformulate the problem, albeit not so that it becomes linear in the optimization variables. However, it is still an improvement speed wise, because the function is nearly linear. We still need iterative optimization, but it will converge much faster than using the original problem. After optimizing the reformulated problem, one might use the result as the initial starting point for optimizing the original problem, thus fine tuning the results.

# What do I still want to know.

Well, the original question is still completely open...

• Interresting regression problem, but not very complicated. Commented Jan 25, 2020 at 22:53
• @Marked42. Your re-edition of the question changes a lot what was asked at first There was no question of speed. OK. now you want $fast$ calculus. But what is your definition of $fast$ ? For example I tested my proposed method with big data (up to 100000 points) The time for result was still not measurable visually. (less than one second). But I supose that is not sufficient for you. Even re-edited, your question is missing of context in order to evaluate the scale of the problem and if it worth the effort. Commented Jan 29, 2020 at 21:59
• @JJacquelin: No, the question stays the same. Just the motivation was added. The question has been all along: "I am looking for a function that has roughly the form of ... but is linear in its model parameter." The fact what I want to do with it was just a motivational add-on. The fact that it is about speed is also only a motivation behind the motivation: I want to use linear least-squares is what I wrote. The fact that it is about speed does not change the question. Any discussion of regression and optimization and so on is nice to know, but not essential to the question. Commented Jan 29, 2020 at 22:04
• @Marked42. If it is your way to answer, Bye-bye and good luck ! Commented Jan 29, 2020 at 22:09
• @JJacquelin: Sorry, if I offended you. I tried to write factual - English is not my mother tongue. You were criticizing my question twice (first with "not very complicated" and then that I completely changed the question - which is fine - but am I not allowed to justify myself?), but I am not quite sure what you are aiming at: Do you want to know the entire background of the question? I mean like the setting in which the function shall be used? Commented Jan 29, 2020 at 22:13

$$y(x)=\exp\left(−(x/a)^2\right) − \log\left((x/a)^2+1\right)$$ Let $$t=\frac{x}{a}$$ $$y(t)=\exp\left(−t^2\right) − \log\left(t^2+1\right)$$ Consider the inverse function $$\quad t(y)$$ $$t(y)=\frac{x(y)}{a}\quad\implies\quad x(y)=a\:t(y)$$ This is a linear function wrt $$a$$. You can use linear least squares to estimate the parameter $$a$$.

More concretely :

With the data $$(x_1,y_1)\:,\:(x_2,y_2)\:,\:...\:,\:(x_k,y_k)\:,\:...\:,\:(x_n,y_n)$$

For each $$(x_k,y_k)$$ compute the root $$t_k$$ of the equation :

$$\exp\left(−(t_k)^2\right) − \log\left((t_k)^2+1\right)-y_k=0$$

You obtain a new data : $$(t_1,x_1)\:,\:(t_2,x_2)\:,\:...\:,\:(t_k,x_k)\:,\:...\:,\:(t_n,x_n)$$

Proceed to the linear regeression for $$a$$ : $$x_k=t_k\:a+\epsilon_k$$

NUMERICAL EXAMPLE :

• If were happy to solve the question using a general solver, I could have just used a non-linear least squares solver on $min_a \sum_i (y_i - f(x_i))^2$ directly. I guess your transformation adds an advantage? What is the advantage? Commented Jan 27, 2020 at 17:01
• Because of that I asked for a function that looks similar to mine, but which could be solved directly with linear algebra, instead of using numerical iterative optimization methods. I want a result fast. Commented Jan 27, 2020 at 17:16
• The advantage is only to answer to your question which was : " to use linear least squares to determine the parameter " In the proposed method the regression is linear. This answers to the question. Of course this is not an advantage compared to a non-linear regression. I would like to cite the sentence from Claude Leibovici : "I recommend that you polish the solution using nonlinear regression since what is measured is y and not any transform." But don't forger that non-linear regression requires initial guess for the parameter which is a drawback compared to linear regression. Commented Jan 27, 2020 at 17:18
• One more time, this is impressive. I used your data for a full nonlinear regression and got the same result ! Cheers, Jean ! Commented Jan 29, 2020 at 10:22
• I am not sure whether we are using different terminology here. I want to do non-linear regression. This does not work without iterative optimization, which I do not want to use due to low speed. You did not eliminate the need to use iterative optimization (that is what your function "root" basically does). I tried to clarify in my question what I mean. (also @ClaudeLeibovici ) I also added why I believe your answers still contribute to a solution. Commented Jan 29, 2020 at 21:23

To continue from @JJacquelin's answer you need to solve for $$t$$ the equation $$y=\exp\left(−t^2\right) − \log\left(t^2+1\right)$$ Let $$t^2=z$$ and the problem is to solve for $$z$$ $$y=\exp\left(−z\right) − \log\left(z+1\right)$$ Now, assuming $$y>0$$, consider that you look for the zero of function $$f(z)=z+\log (y+\log (z+1))$$ which is close to linearity (this is very good for Newton method. When you have solved the equation for $$y_n$$, use the result as the starting guess for the next point (assuming that you did sort the point by $$y$$).

The iterates will be given by $$z_{n+1}=z_n-\frac{\log (y+\log (z_n+1))+z_n}{1+\frac{1}{(z_n+1) (y+\log (z_n+1))}}$$

If I may suggest, this procedure giving $$z_n$$, then $$t_n$$ and the linear regression $$a$$, I recommend that you polish the solution using nonlinear regression since what is measured is $$y$$ and not any transform.

• Hi Claude ! Well done. I am not so brave as you. To solve the equation I simply used the function "root" of a software. Commented Jan 26, 2020 at 12:29
• @JJacquelin. Hi Jean ! I cannot resist when I see an equation ! Cheers :-) Commented Jan 26, 2020 at 14:22
• I see what you do, but how is this advantageous compared to using Newton's method on my original minimization problem and just directly solve $min_a \sum_i (y_i - f(x_i))^2$? Commented Jan 27, 2020 at 17:15
• Because of that I asked for a function that looks similar to mine, but which could be solved directly with linear algebra, instead of using numerical iterative optimization methods. I want a result fast. Commented Jan 27, 2020 at 17:16