Machine learning in trading: theory, models, practice and algo-trading - page 1055
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
Maxim, once again
=====
1) he takes ordinary extrema
2) He looks for certain structures (levels) through selection of different variants (possibly μua).
3) he then trains a set
====
You jump from the first point to the third, and the second is the most important.
Selection of variants is the enumeration of levels of eps, mgua is part of the NS. So he found 200 levels in increments of 1 from 2 out of 1000 options. What kind of selection could there be?
selection of variants is the enumeration of levels of eps, mgua this is part of the NS. Well he found 200 levels with a step of 1 from 2 of 1000 variants. What kind of selection can there be?
There 1-2, there 1-2, there 1-2, connect and train on this, not on raw data
there 1-2, there 1-2, there 1-2, connect it and train on it, not on the raw data
so let's connect, what's the problem
So let's connect, what's the problem?
No problem) Just note that the signs from step 2) should be repeated on the history and be already essentially working before they get into the network in step 3
This all needs to be formalized, it looks like a separate algorithmNo problem) Just note that the signs from paragraph 2) must be repeated on the history and be already essentially working before they hit the net in paragraph 3
why? how will i know they are working until i do point 3
Why? How do I know they're working until I do step 3?
When they hit the net you won't know anything...
the network has a target - there are different targets but the target is good at predicting and trading
You are asking the network to trade 100% of your data and what if it can make forecasts but only 2% of that data. What do you get out of it?
You need to pick these 2% by bits and pieces from all predictors, at least 30% and send them to the net, not in the form of prices or those silly indicators, but only what is really important.
When they get to the network, you won't understand anything...
The network has a target - there are different targets but the target is good at predicting-trading.
You make the network trade 100% of your data and what if it can make forecasts but only 2% of that data. What do you get out of it?
You need to pick these 2% from all predictors, at least from 30% and send them to the net not in the form of prices or stupid indicators, but only what is really important.
yes right now, almost done
A year ago I did some awesome experiments:
1. I took a tick BP with some sliding sample volume
2. I counted the averaged classical variance and averaged spread (High-Low). I.e. at every new tick I recalculated these values, recorded them into separate arrays and calculated the averages in these arrays.
3. surprisingly, these average values coincided with the multimillion tick data
4. The price was sort of moving between those average levels.
5. Then I came here to the forum and got pissed off with everyone else...
Where am I going with this?
Maybe we should work with spread (High-Low), rather than with separate extrema?
Variant 1: without transformation of traits. Training from 08.01 All that before OOS
Maybe work with a range (High-Low) rather than individual extrema?
An extremum is an extremum
(High-Low is volatility
This is not opposed