Machine learning in trading: theory, models, practice and algo-trading - page 2836
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
No, unfortunately.
I'm looking at Lopez de Prado's website now . https://quantresearch.org/Patents.htm.
He has a new patent, issued in September (Tactical Investment Algorithms through Monte Carlo Backtesting).
Many valuable ideas, for example he emphasises nowcasting (short-term forecasting).
Quote: "Shortrange predictions are statistically more reliable than longrange predictions".
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3562025
Key findings from the coronavirus pandemic.
What lessons can we learn amid this crisis?
1. more nowcasting , less forecasting
2. Develop theories, not trading rules
3. Avoid allregime strategies
No, unfortunately.
Did you have a good training or no training at all?
Did the training go well or no training at all?
Normal, in the sense that the calculation is correct (I had max Sharpe at level 3 - 4), but since the class is without njit decorator, the training is very slow.
It's ok, in the sense that the calculation is correct (max Sharpe I had was at level 3 - 4), but since the class is without the njit decorator, the learning is very slow
What is a njit decorator ?
What is a njit decorator ?
an attachment from pandas developers to speed up code execution in functions
It doesn't matter if it's in history or in the future. And the tester itself has nothing to do with it.
What is important is the property of the algorithm (optimisation algorithm individually or as part of a grid) to find the global optimum of the evaluation criterion. I emphasise - the evaluation criterion. The evaluation criterion is not necessarily and/or only profit. It can be anything, for example, the evaluation criterion of work on OOS is not a criterion (minimising the difference between sample and OOS)? - it's just a thought. The criteria can be anything and of any complexity. It is important to understand that the "Profit" criterion is a very gullied and discrete thing, so people try to come up with smoother, more monotonous evaluation criteria, which in general improves the quality of optimisation itself and neuronics training in particular.
Therefore, coming back to what I have drawn on the highly artistic picture - a visual illustration of the fact that in conditions when neither the number nor the characteristics of local extrema are known, the only way out is to search as far as possible for the one that is at all possible under conditions of limited computational capabilities.
Plateau - yes, there is such a notion, but it is not related to optimisation, it is a question of classifying similar sets of parameters by some attribute. Looking for a stable plateau is a separate complex task.
Once again: extrema are of no value: an unstable point, which also does not exist, since we are dealing with random processes, and non-stationary ones at that.
We need to look for a plateau, even the one shown in the figure, as long as it is profitable, even if it is above the local and global minimum. Such a plateau will theoretically show the upper boundary of profitability of the TS. And the found extrema are nothing at all - they are definitely not in the future, but there is hope for a plateau
You have a strange idea of a "plateau". a plateau is not a type of area on the function under study. a plateau is a set of sets of parameters on some attribute.
Profit has nothing to do with non-stationarity of series and even less with the ability of the network to learn and work with similar results on unknown data.
you take a function like profit literally, i.e. as if this function looks like a balance graph in a tester. this is fundamentally wrong.
Once again: extrema are of no value: an unstable point, which, moreover, does not exist, since we are dealing with random processes, and non-stationary ones at that.
We need to look for a plateau, even the one shown in the figure, as long as it is profitable, even if it is above the local and global minimum. Such a plateau will theoretically show the upper boundary of profitability of the TS. And the found extrema are nothing at all - they are definitely not in the future, but there is hope for a plateau
Optimisation is not about finding something you don't know what, but about improving the profitability of the TS, for example.
You have the whole surface of interest in a normal situation, and especially the extrema as an increase in potential profit, let's say. And you will have time to roll down to the plateau.
Or the whole surface has sunk and there are only extremes and plateaus sticking out of the water that need to be found? Then it's already an initial fit.
It's like getting together and saying, "Let's optimise something and look for a plateau", and if we find a plateau, there's a treasure there. Is that how you see it?
Of course there is hope for salvation, but it is negligible.
I imagine the process of finding working sets of neural networks to be illustrated by an example: there is an imaginary mountain Everest, which has one single maximum. the task is to lay a road to the maximum so that the angle of the road does not exceed 3 degrees and the path should be as short as possible. If you search for a straight path to the top, then a lot of people will surely crash and die (they will cry when they land at the bottom, flattened, complaining about the non-stationarity of the wind rose). why do you need such a maximum? there is another function, the derivative of the mountain function, which includes two criteria - the angle of inclination of the mountain and the length of the road. This problem is reduced to optimisation and finding the minimum of this derivative function from the mountain (minimisation of the inclination angle and road length). so this function will have many local extrema and only one global one. thus solving this problem we will find a safe way to the top of Everest, no one will die and the non-stationarity of winds will not matter to us. but "plateau" should be understood not as a certain platform on the mountain where you can take a break, but as a group of path options with similar characteristics of angle*length satisfying safety requirements.
That's how it is, in a general way. Not every algorithm is able to find a "safe" road. both search properties and convergence and convergence speed are important.
Further, why do they introduce derivatives of a function to solve problems like the example above? To make the surface of the function smoother.
Another example, quite close to practice. there is a problem with 10 questions, the network must answer 10 questions. let's count the number of answers. the maximum in this case will be 10. that is, the function of this problem is discrete. with many local extrema. why? let's look here:
Let's count the right and wrong answers. 0 is wrong, 1 is right.
0 1 1 0 0 0 0 0 1 0 0 0 1 1 1 : 4 correct.
1 1 1 0 0 0 0 0 0 0 0 0 0 0 1 : 4 correct.
0 1 0 1 0 1 0 0 0 1 1 0 0 0 : 4 correct
0 0 0 0 0 0 1 0 0 0 1 1 1 1 : 4 correct
0 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0 0 : 4 correct
.......................................
we can see that the network gives answer choices with a score of 4. this is an example of how the network is stuck. it does not see further growth. shuffling answers cannot score more points.
how to make the network more efficient? right, to do this, the original answer function is converted into an error function. so, minimising the total error over all answers, the network will find all the correct answers. in this case, the problem is reduced to minimising the error, the minimum is global 0, the function has many extrema, but now it is smooth, the game is reduced to cold, warmer, hot, hot, which is much "smoother" that just yes or no.
but. unfortunately, the problem function cannot always be transformed into a smooth function. the presented functions in articles and should be taken as already transformed, while the original ones have a much more complex topography. thus the efficiency of optimisation algorithms lies in the ability to find the extremum of the derivative of the function.