Machine learning in trading: theory, models, practice and algo-trading - page 1008
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
As an alternative, you can use the discretization. We break the series by conditions and analyze the broken parts as to whether they belong to a pattern. That is, we break the problem into parts: first the elementary units, then their totality.
In principle, this is automatically done in convolutions, as Maxim suggested. The convolutions can be done by row of numbers, or by screenshots or generated images from the time series.
But this requires a little understanding of the subject. The patterns are in layers. Also from primitives to more complex abstractions. If really interested, there is a base for research, I can also give direction or help with code.
Do you mean to decompose the image into pixels, like letter recognition in machine learning? My search is simply based on the price series of two arrays.
If it is possible to do somehow to find the pattern you are searching for with fewer bars, for example the standard pattern is 150 bars, and a similar one is found in the history but with the size of 70 bars,
it's interesting.
What is shown in the picture are the patterns found by the algorithm (script).
I don't really understand your definition of Markovism, but it doesn't seem to be quite the same as the usual one. For example, a trend (as in your picture) and a Markovian are quite compatible.
The probability of such a series in such a sequence (even in the case of symmetric wandering) is higher (a problem from the field of elementary combinatorics).
With random walk the mathematical expectation is equal to the initial value; but in Forex there is also a spread that will decrease funds on each new trade.
Do this experiment - flip a coin and draw a random walk graph, if heads fell out then the graph is still up, if tails then still down. You get a random walk as usual.
Now do the experiment again, but now with each flip of a coin, draw the graph down a little more. With a large number of tosses, there will be about as many upward movements as there are downward, but there will also be a lot of small downward movements. And these small downward moves will slowly but surely steer the entire chart downward. The mathematical expectation will not be equal to the initial value.
So if the chart should be tending downward at about -4 cents per trade, but for some reason it isn't - it means something is wrong with the random walk and Markov's process.
There is a problem that 5 weeks is not enough time to evaluate the signal, a lucky person could also get such a signal by trading a coin, you have to wait a couple more months for credibility.
Doc, you're not a role model.
Show me a 2-week work.
Shame on you. This is just a demo of what some have not tried, but still stubbornly called impossible.
I will not have a better signal, I'm done with forex. Someday crypto exchanges with cryptocurrencies will appear, I'll show something there.
Once again, well done, Doc.
In fact, this is the first sustained neurosignal I've seen. Even if it's on a demo.
So there you go, you miserable old men sitting here on the forum, instead of playing dominoes in the yard! Learn how to work!
I won't have a better signal, I'm done with forex. Someday crypto exchanges with cryptocurrencies will appear, I will show something there.
Cryptocurrencies can collapse like the Nasdaq Index did in the 2000s and then many dot-com stocks that were part of that index never rose to those heights. Same analogy.
But, as long as cryptocurrencies are trending, and it is easier to make money on a trend, it is an axiom. All fortunes in trading were mostly earned on trends.
Maybe, it would be better to try stock trading, there are also trends there and it is not so dangerous. When cryptocurrencies collapse, many of them will just lose their value.
Now I look, and bitcoin is already at half of its absolute maximum and the volatility is huge.
And in general it will be interesting to see your signal, how all this event will end)
With random walk the expectation is equal to the initial value; but in forex there is also a spread that will decrease funds on each new trade.
Do this experiment - flip a coin and draw a graph of a random walk, if heads roll, the graph is still up, if tails roll, then still down. You get a random walk as usual.
Now do the experiment again, but now with each flip of a coin, draw the graph down a little more. With a large number of tosses, there will be about as many upward movements as there are downward, but there will also be many small downward movements. And these small downward moves will slowly but surely steer the entire chart downward. The mathematical expectation cannot be equal to the initial value.
So if the chart should be tending downward at a rate of about -4 cents for each trade, but for some reason it is not - it means something does not coincide with a random walk and a Markov process.
Having a trend (slightly down because of the spread) doesn't make random wandering non-markovian. I think you're confusing markovality with the property of being a martingale - if a random walk has a downward trend, it will already be a supermartingale instead of a martingale.
That's the easiest thing is to show the signal from the demo.
Nothing complicated, I just did it.
Thank you!
I've been wanting to ask a question for a long time. Suppose I have a head and shoulders pattern 150 bars long. I need to find similar patterns by history, but they will be found if the number of bars is almost the same in the pattern itself and in the pattern found. How to get away from the exact number of bars look for some frequent one and derive an average for it or something else?
Try the DTW method - that might be what you need.
As for sampling.
It is possible to divide the entire time radius, for example, by a fine zigzag. Then sections from "pit" to "peak", and vice versa, will be "elementary units" of the pattern (upward movement, downward movement, small upward movement, etc.). The analogy is letters.
Accordingly, we further analyze them in pairs and threes. For example, "here is the likeness of the left shoulder," "here is the likeness of the head looking down." The analogy is syllables.
Then it's just a matter of looking at the sequences of "syllables" and checking that they belong to "words".
As you can guess, the sample size will no longer matter, only the semantics.
I hope I made myself clear.
Regarding convolution.
The convolutional layers can be applied to various data, not only images. They are now successfully applied to text and sound. So, their potential in the market has not yet been explored (at least openly).
It makes perfect sense. I even read about it somewhere.
Once again, well done, Doc.
In fact, this is the first sustained neurosignal I've seen. Even if it's on a demo.
So there you go, you miserable old men sitting here on the forum, instead of playing dominoes in the yard! Learn to work!
You can't not crap, but the dirt gets around...
You can't not shit, and the dirt spreads...
I have no time to bicker, for I am not young myself. Especially since I just caught the Grail.
My friends!
The money will soon be pouring into our pockets like a full-flowing river!
Uncle Sasha did it, as promised.