Machine learning in trading: theory, models, practice and algo-trading - page 1570
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
I used to just do it in increments, and the result was zero. I even recorded a few videos on this topic
I added time cycles (hours, days, months, etc.) to the gradients and it started to work (so far without MO). I.e. dependences lie in time intervals, which interact with each other and each with itself, but with a lag.
Hours, days, etc. are categorical features that can't be compared to each other, but there are models in which they fit well, such as CatBoost. Plans are in the works to try it.
Basically, this is the logic block - the interaction of different time slots with each other (see my last article). I think MO should be able to handle it, too. No other thoughts as to what other patterns there might be, other than lag dependencies.
I'll read articles and as for thoughts - there is a hypothesis about so called self-fulfilling technical analysis, i.e. Technical analysis works because most traders believe that it works. This also applies to RA and other relatively new chips.
I want to use all their variations and teach the neural network to make predictions based on all their signals.
I added time cycles (hours, days, months, etc.) to the increments. I.e. dependences lie in time intervals, which interact with each other and each with itself, but with a lag.
What period of data did you use?
I remember training on 3 or 4 months of M1. And the forest found purely on time as an input, that on Tuesdays from 9:10 to 9:30 you have to buy and almost always got TP, not SL.
But after shifting the learning interval, this pattern disappeared.
Still, the interval should be at least 1-2 years.
What period of time was the data used?
I remember training on 3 or 4 months of M1. And the forest found purely on time as an input, that on Tuesdays from 9:10 to 9:30 you have to buy and almost always got TP, not SL.
But after shifting the learning interval, this pattern disappeared.
Still, the interval should be at least 1-2 years.
10 years
Thanks, I will read the articles, and as for thoughts - there is a hypothesis about so-called self-fulfilling technical analysis, i.e. Technical analysis works because most traders believe that it works. This also applies to RA and other relatively new chips.
And if we collect all their variations and teach the neural network to make forecasts based on all their signals.
I don't know, I think that these are myths.
The eternal unsolved topic is if a coin goes on tails 10 times, what is the probability of heads. In a perfect mathematical model, it's the same = 50. In the real world... if I'm not mistaken, no one can prove or disprove whether the tosses depend on each other.
A coin has no memory is one of its definitions. The probability of an eagle/dash is always a constant, no matter what the story was. A coin doesn't know history, so it doesn't matter how many times in a row a coin rolls tails, 10 or 20, the probability of tails is strictly 50/50, as long as the coin is "fair.
Therefore, if the ticks of a random walk are generated by a coin, then such a walk also has no memory and atANY point of the trajectory of the accumulated amount, the probability of further movement up or down is absolutely the same. Accordingly, the probability of winning and losing on a random walk, in the absence of a spread, is exactly the same.A coin has no memory-that is one of its definitions. The probability of heads/tails is always a constant, no matter what the history was. A coin doesn't know history, so it doesn't matter how many times in a row the tails fall, 10 or 20, the next flip is strictly 50/50 if the coin is 'fair'.
Therefore, if the ticks of a random walk are generated by a coin, then such a walk also has no memory and at ANY point in the trajectory of the accumulated sum, the probability of further movement up or down is exactly the same. Accordingly, the probability of winning and losing on a random walk, in the absence of a spread, is exactly the same.Prove this hypothetical nonsense. Not from books, but to be clear.
Everyone here is very good at quoting wikipedia. Where's the practice?A coin has no memory-that is one of its definitions. The probability of heads/tails is always a constant, no matter what the history was. A coin doesn't know history, so it doesn't matter how many times in a row you get tails - 10 or 20 - the next flip is strictly 50/50 if the coin is 'fair'.
Therefore, if random walk ticks are generated by a coin, such a walk also has no memory and at ANY point of the accumulated sum trajectory, the probability of further movement up or down is exactly the same. Accordingly, the probability of winning and losing on a random walk, in the absence of a spread, is exactly the same.So you're saying that the correct strategy after 9 tails is 50/50. But what about the overall probability of tails (9, 10... times in a row) - you can't ignore it, even if the coin is fair.
After all, if we do a model experiment with a fair coin. Then in cases where we measure the probability only after 9 tails, in that case the probability will not be 50/50. Experiments rule. A coin doesn't have memory, but the universe does ))))
And you prove this hypothetical nonsense. Not from books, but to make it clear to yourself.
Everyone here is eager to quote wikipedia. Where is the practice?Prove what exactly? That a coin has no memory? - So this is its definition.
Or the fact that the probability of the outcome after any previous series is always 50/50 follows from the fact that the coin has no memory.
So you're saying that the correct strategy after 9 tails is 50/50. But what about the total probability of tails (9, 10... times in a row) - you can't ignore it, even if the coin is fair.
After all, if we do a model experiment with a fair coin. Then in cases where we measure the probability only after 9 tails, in that case the probability will not be 50/50. Experiments rule. A coin has no memory, but the universe does )))
The probability of OOOOOOOOOOOOOOOOOOOOO is the same as the probability of OOOOOOOOOOOOOOOOO or OOOOOOOOOOOOOOOOOOO.
The probability of falling out LLCOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO is the same as the probability of falling out LLCOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO.
And you prove this hypothetical nonsense. Not from books, but to make it clear to yourself.
Where's the practice?Well, it's easy - even you will understand it.
Generate ANY row with PRNG, pull any TC on it, as the great trachter builder does, get a positive result - pulled it well.
Then generate another 50 rows with the same PRNG and apply the same TS to all of them with the same settings as the first time, and get a resulting pooch or close to 50/50.
If you need more caca, you generate a lot of rows.