Machine learning in trading: theory, models, practice and algo-trading - page 241
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
Tried it, it didn't work. It is possible to pick up clusters for quite a nice trade on sample, but on oos it is almost always a drain, bad strategy.
What exactly was clustered?
and what was the target?
I liked the article just fine.
I mean the whole series of posts about candles.
Sanych well take it and do it!!!
1) tell the gist of the idea
2) write and post the code
3) show pictures of trade on the oos
just blah, blah, blah...
Make at least one normal post, which can not in words but to specific studies to confirm your point of view, which you so fiercely promote, and that it was possible to touch, but you do not do it, right? I know why.
Well, you're wrong.
Give me the target and the predictors, and I'll give you back an estimate of their importance to the target. I have a few examples (I think I did more than 10 on order), but the result is always the same: the model did not retrain on the selected predictors. And the classification error is determined by the list of predictors. If the error is large, you need to change the predictor list.
Waiting. Better if .RData. This format is quite available to you.
By the way, I offer it to everyone. Free of charge.
Well, that's not true.
Give me the target and the predictors, and I'll give you back an estimate of their importance to the target. I have few examples (I think I did more than 10 on order), but the result is always the same: the model was not retrained on the selected predictors. And the classification error is determined by the list of predictors. If the error is large, you need to change the predictor list.
Waiting. Better if .RData. This format is quite available to you.
By the way, I offer it to everyone. Free of charge.
What does this have to do with what I wrote?
Eh, Sanych, Sanych...
What does this have to do with what I wrote?
Eh, Sanych, Sanych...
1. That I'm not posting anything. You're wrong.
2. I am ready to apply my skills and algorithms to other people's data to confirm the thoughts I have repeatedly stated.
3. and regarding some of your posts, as well as some of your other posts
You are trying to form predictors. Yes, this is a problem, and a problem from the beginning and all sorts of thoughts on the subject are of interest.
But.
A bunch of posts where the formation of predictors has poured out for the sake of the formation itself.
First, the aim of the research is defined, and then the research itself. Can it be any other way?
Hilarious:
"Asked the forum what was wrong with my code. Learned a lot about myself."
The posts weren't about candles...
There is a persistent smell of shamanism from all these exercises with candles. There is no regular thought at all! Just an amazing composite of high-brow machine learning algorithms and typical technical analysis nonsense.
The whole point is in patterns, when several candlesticks in a row form a pattern. It really works, but such patterns are very difficult to find. For example, the pattern strategy that I know works on D1 for some pairs and you need a lot more than two candles and when the pattern coincides, you need not just open a position, but you need to put a pause at a certain level and then trailing by a special technique. All this is not very profitable. I do not like D1 and it is difficult to search for alternative patterns by oneself. But still it is profitable.
What exactly was the clustering?
What was the target?
It's all about patterns, when several candles in a row form a kind of pattern. It really works, but such patterns are very hard to find. For example, the strategy with patterns that I know works on D1 on some pairs, and you need a lot more than two candles, and when the pattern coincides, you need not just open a position, but you need to put a pause at a certain level, and then trailing by a special technique. All this is not very profitable. I do not like D1 and it is difficult to search for alternative patterns by oneself. But still it is profitable.
I took candlesticks, made 4 predictors from them as described by _Vizard, trained som, found a cluster number for each line in the table and found the most profitable action with each cluster (buy/sell/exit) by method of selection. There is no goal, there is a selection of actions according to the number of the obtained cluster.My outburst of emotion is not related to the candlesticks, as one might get the impression.
It's about something else.
When it comes to candlesticks, many books have been written that list numerous combinations of candlesticks. Moreover, there are people who claim to be able to trade profitably.
My complaint with all of these candlesticks in particular, and with technical analysis in general, is that profitability in the past has NOTHING to do with the future. And if we take models, which are much more complex than the "three soldiers", we should get something in return. And to my mind, this is proof that the future will be like the past.
If we get into R, or more specifically machine learning, mathematical statistics, it's only for the sake of:
1. For regression models, we build predictions on stationary predictors
2. For classification models, we know how to shield them from overtraining, at least by prefiltering the noise predictors.
If we know how to do such things, then we have a tool that is qualitatively different from technical analysis. If we can't, we don't have to bother.