Machine learning in trading: theory, models, practice and algo-trading - page 1282

 
Aleksey Vyazmikin:

Imagine the predictor is day/night, and you will have, for example, more than 1 in your targets during the day, is this a good predictor? Or it's not about the day, it's about the fact that important news (affecting the market) comes out during the day more often than at night.

I don't think it's appropriate to judge a man who does a lot to promote MO here...

Of course day/night or otherwise formalized time is an important predictor, as confirmed by TS built with trading sessions in mind.

And of course I'm not talking about those who do and write a lot, but who just blah-blah-blah:)

 
Kesha Rutov:

I am OK, stable 10-15% error, on tests. On the real, everything is still confused, uncertain, but I'm trading, taking risks, unlike you and similar students of studs, sitting on the neck of elderly parents.

Unfortunately (or fortunately) we don't know anything about your trading, just like everything else

so for now you're just a nobody expressing your opinion which no one wants

that's reality

If you just want to get high on the topic that me and the guys raise, then at least write interesting things instead of bird babbling.

 
Maxim Dmitrievsky:

Unfortunately (or fortunately) we don't know anything about your trade/drawing as well as everything else

so for now you're just a nobody expressing an opinion that nobody wants

that's reality

If you just want to get high on the topic that me and the guys are raising, then at least write interesting things instead of just babbling.

+

 
Maxim Dmitrievsky:

Unfortunately (or fortunately) we don't know anything about your trade/drawing as well as everything else

so for now you're just a nobody expressing an opinion that nobody wants

that's reality

if you just want to get high on a topic that me and the guys are raising, then at least write interesting things instead of just babbling.

on the hippodrome

 
Kesha Root:


Kesha, can you stop fooling around and at least say something on the subject? I assure you, the Zigzags have nothing to do with BP analysis. You can throw out all the articles on the subject.

 
elibrarius:

I limited the tree to 1 line:

samples++; if(samples < 20){ then don't split the node anymore, but leave the leaf}

i.e. there will be at least 20 examples left in the sheet, for representativeness.

That's all the release you asked for)))

This is already good, but I would raise the limit to 50-100 examples, and the less predictors, the greater this value should be in my mind. As for the release, so it's better to immediately code, not everyone understand the structure of the library and know what and where to fix.

elibrarius:

The degree of understudy, ie, the number of examples in the sheet can be any 10, 100, 1000 or optimize. In xgboost it is called min_child_weight

Then what stops branching? By the way, you can add the maximum number of splits to the release then ;)

 
Alexander_K2:

The zigzags are in no way relevant to BP analysis.

Controversial statement.
We should make them non-seeking, i.e. count from the 0th bar forward and/or backward. Wizard seemed to say.

 
Aleksey Vyazmikin:

This is already good, but I would raise the limit to 50-100 examples, and the less predictors, the greater this value should be in my mind. As for the release, so it's better to immediately code, not everyone knew the structure of the library and know what and where to fix.

Then what stops branching? By the way, you can add to the release and the maximum number of splits then).

I will not do any release for one string. I understand that someone else's time is less valuable than my own. And I take exactly the same position.
I showed a simple solution. Whoever needs it, will repeat it.
 
Ivan Negreshniy:

Of course - the day/night or otherwise formalized time is an important predictor, as confirmed by the TS built with the trading sessions.

And of course I'm not talking about those who do and write a lot, but who just blah-blah-blah:)

I work with time myself, it's a matter of logical reasoning, at the expense of more significant predictors (news) than just the time of day.

I don't understand who you are actively alluding to, and I'm not even interested, since discussing personas is not a meaningful activity for this thread.

Write your ideas on the subject, if you have any.

 
elibrarius:

That's a controversial statement.
We should make them non-seeking, i.e. count from the 0th bar forward and/or backward. Wizard seemed to say.

Absolutely all TCs that exclude such a concept as "time" are all doomed to failure. This is one of the secrets of the market. Just - shh... not a word to anyone!