Machine learning in trading: theory, models, practice and algo-trading - page 3392
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
Who said anything about consistency?
I was picking on the thesis because I just wanted to speculate.
Yes, I'm aware of that.
I'm curious about the practicalities. What it gives.
For example, a time sequence can be characterised as follows: until you finish at the current step, you cannot move to the next one. Until you convert the current time value, you can't move on to the next one. And you use the results of the previous step in the next one.
As in NS layers.
And in the sequence - the convolution goes in any direction: from left to right or from right to left, it doesn't care. It still summarises everything, but it puts the data in the right order.
This is probably an example.
Having a sequence, it doesn't matter at what point in time each element was obtained. They can be drawn arbitrarily. The last one can be drawn first and so on, if at all on fingers. And BP elements are ordered by the time of their appearance.
And all of the above is also true.Who said anything about consistency?
We all lack some kind of trial dataset like iris or mnist but with market data to test and compare the performance of our ideas and AMO
.
Golden words! Only to create such a trial dataset, you need to have at least a rough model of pricing in the market. Like - how does the price formation process happen? Because if we consider that the price is just a random walk, with the genesis of a symmetric coin, then there is nothing to catch - it is easier to have fun in the casino.
Golden words! Only to create such a trial dataset, you need to have at least a rough model of pricing in the market. Like - how does the price formation process take place? Because if we consider that the price is just a random walk, with the genesis of a symmetrical coin, then there is nothing to catch - it is easier to have fun in the casino.
I tried to train (optimise) 10 years on straight dollars. The result of the best set is, of course, a flat (conditionally) line of balance growth.
10 years. That's not a week, not a month. During that time, the price chart has had time to experience long-term upward trends and long-term downward trends. The same thing and medium- and short-term.
But as soon as I tried to switch to reverse dollar pairs - there was a stable even drain.
Switching to crosses - random-fleet up and down.
That is, according to all the canons of retraining in the reverse error propagation, and when optimising in the tester, usually remembering the path goes only for one currency pair (on which you teach), and the others show random. Yes, there are similar results, but it was a random pair, when you optimise on Eurodollar and the network shows a profit on some francoyenne.
But, all dollar pairs have no correlation under 1. And the result is almost the same profit on the training period for the direct dollar ones, and exactly the same with a minus sign for the reverse ones.
So, the conclusion is self-evident: pricing is not a random wandering, but an archically complex system.
We all lack some kind of trial dataset like iris or mnist but with market data to test and compare the performance of our ideas and AMO
.
It will turn out exactly the same way as with the test f-iels - fitting to a particular set. You take Eurobucks and show what you've done. If you've made a good one, you put it on the market. And send it to us in ONNX format.
So the conclusion is self-evident: pricing is not random wandering, but an archly complex systemic one.
The TS is simply fixed on the general trend. I often get a similar picture on correlated instruments.
Especially if the signs are invariant to the volatility of each instrument.You'll get exactly the same as with test functions - fitting to a specific set. You take Eurobucks and show what you've done. If you've made a great one, put it to trade. And send it to us in ONNX format.
Everything is cool except for the last one, you can't put complex code into ONNX, except for the ready model.
You probably won't even know what I'm talking about.
If there was a docker container, then yes, there are no limitations, but with ONNX it's one big limitation.Everything is cool except the last one, you can't put complex code into ONNX except for the finished model.
You probably won't even know what I'm talking about.
If there was a docker container, then yes, there are no limitations, but with ONNX it's one big limitation.