Machine learning in trading: theory, models, practice and algo-trading - page 747
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
Well, this information is what we get from all sorts of possible input parameters. I started out with just wands, then with incremental wands, then with delta wands... Now I'm working on something like a borscht of turkeys )))) in order to feed one input line instead of 20, like before...
Just on the borsch ))) all the turkeys are derived from the price. For example at the close of yesterday's candle indyuki showed a result such as stochastic 14 7 4 (taken at random) showed 44.44 and 27.78 - these 2 numbers well describe yesterday's day and the next days. But not today... How much information is there in these 2 numbers about today? 0.0001? ))) I think we need to look for other values... Something that is more informative about the future.
I believe that the duration of the backtest period, and only it, can play the role of the judge. If there is no explicit backtesting of deals by dates or their sequences, and there are thousands or tens of thousands of deals over several years with a smooth increase, then it's already good
And what information is not so important.
It is not enough. Let's not forget that people who preached the hypothesis of an efficient market worked successfully for years, got nominees, and then went bankrupt.
The backtest is necessary, preferably in the tester, as the closest to reality.
BUT.
There must be a theoretical basis for the success of the backtest.
With this you can have confidence in the backtest.
Max, in idea you want to teach the machine to recognize different market phases to automatically select the inputs that will be the most effective for each state. It's like a portfolio of several neural networks, where each of them is trained for a certain market condition...
Yes, something like this, but you have to introduce all sorts of global meta-states that affect groups of TC parameters, and sub-states, like the Markov
Just for borscht)) all indices are derived from the price. For example at the close of yesterday's candle indyuki showed a result, such as stochastics with parameters 14 7 4 (taken at random) showed 44.44 and 27.78 - these 2 numbers well describe yesterday's day and the next days. But not today... How much information is there in these 2 numbers about today? 0.0001? ))) I think we need to look for other values... something with more information about the future.
When a trader looks at a chart, he sees something and then he needs to describe it by a certain rule. sees where the price has been for a long time, sees where it's accelerating... And that's without indicators, but you can add indicators and it increases image comprehension by times.
I tried it with pure increments, but I couldn't get anything out of it... I must have set the target wrong... can you tell me?
The input of the neural network should be the sum of these purest increments in a certain observation time window.
ALL.
The input of the neural network must be the sum of these purest increments in a certain observation time window.
ALL.
Okay, I'll try to make different time windows of sums, only I don't have a grid, I have a forest, but I think it doesn't matter if the algorithm is different. on the contrary, you can put several time windows in the forest at once.
This is not enough. Let's not forget that people who preached the hypothesis of an efficient market worked successfully for years, got nominees, and then went bankrupt.
The backtest is mandatory, preferably in a tester, as the closest to reality.
BUT.
There must be a theoretical basis for the success of the backtest.
With this in place, one would be able to trust the backtest.
Yes, but my system operates on the concepts of Holy Spirit and macaroni monster, so it is very difficult to explain theoretically
For example, a trader sees the yen range for the time that is displayed on the screen, sees where the price is relative to the range... and then the imagination begins to help. sees where the price has been for a long time, sees where it's accelerating... And that's without indicators, but you can add indicators and the image comprehension increases many times.
Exactly!!! A human trader doesn't see numbers, he sees simpler models. approximated, smoothed, but not by a moving average, but for example by the SSA method (something like that) - but you can't feed it to a neural network...
Yes, but my system operates with the concepts of the Holy Spirit and the macaroni monster, so theoretically it is very difficult to explain
just say it's lazy to prove it )) it's obvious that you are an excellent student lazy ))) it's easier to do than to prove it ))))))
I will try to make different time windows of sums, only I do not have a grid, I have a forest, but I think it does not play a role that the algorithm is different. on the contrary, in the forest you can make several time windows at once.
For the windows, take my basic TS, there is always a different window dictated by the market itself and its formation implies a market reversal. The most normal moment for analysis is the moment of anticipated reversal.
Max, I will wait for your apology for my skepticism about my TS in poems, which I should also like.
My market has long changed from upward to downward and backward, and TS still works. And if you had seen the model itself, you would have been surprised (it's very small).