How to find reversal points in a graph

Question

Edmundas Povilavicius 2017-2-17

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/325630-how-to-find-reversal-points-in-a-graph

编辑： dpb 2017-2-24

Hello,

please help me find subsequent 8 reversal points in a price graph. I have indicated in the image bellow what should be specified as a reversal (highs and lows) after the initially drowned retracement levels (dashed blue lines).

Some other useful info:

Date represents one of the candles, and the rest data points are explained. Output matrix is just a matrix of 8 labeled reversal levels.

Assume that retracements levels for data sets are already computed from the same data based on specific ranges (so all the labeled levels are at prices like 2134, 2098, 2050, etc.). Also, margins for distance between the price low/high and a level will be specified manually.

P.S. I have looked into "findpeaks" function but I don't think it;s robust enough to work.

Thank you in advance!

10 个评论
显示 8更早的评论隐藏 8更早的评论

Star Strider 2017-2-18

The problem is the ambiguous information.

If we have a way to calculate the ‘fluctuation’ variable or set of conditions, and if you provide it and the relevant data as a file rather than an image so we can work with it, and explain the problem and the way you would usually solve it without using MATLAB, we most likely can come up with something that if not exactly what you want, is close enough so that you can tweak it to make it work in your application.

There aren’t many econometricians here, if any. We’re engineers, statisticians, and mathematicians with specific areas of expertise in applied mathematics, image processing, computer science, and for me, clinical and basic science Internal Medicine and biomedical engineering. In any event, please do not assume that if you have specialised knowledge that there are others here who share it. The vast majority of us here are unpaid volunteers who do this for fun, because solving problems are fun.

We are primarily here to solve coding problems and make code more efficient. You will not insult us if you describe any problem in detail, describing terms and concepts. It actually helps us know what you know. With enough information, if necessary we can go online and do further research, or ask you for clarification.

This is ideally a learning opportunity for all of us here.

dpb 2017-2-18

编辑：dpb 2017-2-19

Well, sorry you don't seem to see the aproppriateness of the story; it wasn't meant as a joke at all. (It is, in fact, a true characterization--in that environment since it was my day job I could spend a lot of time looking at and under rocks and eventually come upon the right one (or break off chunks of a boulder to create the right one, more often). ) The manager, while not techically-trained, did have an inate sense for things and we did collaborate on what turned out to be innovation winners on several of these what seemed to be wild-goose chases for a while, simply because he wouldn't let go the nagging feeling that there was something not right. Unfortunately, he's now passed and since all this was at the Oak Ridge DOE site and classified owing to its relationship to the programs there, most of these nuggets will remain buried in the DOE Q files archives. :(

Anyway, now past the old geezer part, as Star points out at some length, we only know what is posted; without the data itself and some really serious attempt at defining what does and what doesn't qualify (and why) other than just a few examples, the likelihood of anybody being able to create a satisfactory algorithm in the level of effort normally indulged in here is essentially, imo, zero. This is, after all, a volunteer group and while all the regulars are pretty "in to" doing neat things w/ Matlab when queries are posted here, there's a limit to just how much time and effort one can devote to the forum and still have a real life.

If this process is something known in econometrics circles, give us the references; if it's something you're just trying to make up on the fly as research or personal interest, still at least show us what you've done to date and where it's not satisfactory and why not...

Edmundas Povilavicius 2017-2-24

Thanks everyone for your time and for helping me out. I realized that this might be too complex and would take too long to describe and find a coding solution. I chose an alternative method - to do it manually. Thank you everyone once again.

dpb 2017-2-24

编辑：dpb 2017-2-24

Sometimes simpler is better... :) What can be essentially trivial to inspect and select by eye can be exceedingly complicated to automate; the eye/brain are quite adept at pattern recognition, zero and one "not so much" at least without quite a lot of help.

Does look an interesting problem, if there's future need to automate this, don't hesitate to post it again, but when do so, throw in some more background info and definitely do not neglect to attach real data.

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Image Analyst 2017-2-19

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/325630-how-to-find-reversal-points-in-a-graph#answer_255428

If findpeaks() is not robust enough then you can try smoothing the data before using it, like with sgolayfilt() or conv(). Or try harder to find the right parameters for findpeaks(). There are a lot of them and , granted, they're confusing. Perhaps you just haven't found the right set of input settings for it yet. Otherwise you might have to write your own code that can handle whatever rare quirky data is not correctly identified by findpeaks().

Perhaps you can include a set of data and pinpoint where in the data findpeaks() is not working correctly.