Machine learning in trading: theory, models, practice and algo-trading - page 1812

 
Aleksey Vyazmikin:

Long is how long - I have a model that has been running at least half a year on the plus side. I discovered it with about a month ago, when I was parsing old file archives - took a model and it works, but now I do not know how it was trained - model CatBoost.

Another example - also built on leaves, most of which are collected on the 2014-2018 sample inclusive, a tree in February 2020, which is used as a filter, and this synergy works well in 2020.

However, I didn't risk putting money on it all - and that's my mistake.

Now it's haunting to think that just as I get to the criterion long and everything breaks.

If you think that the criterion is a real criterion, you may use it as a guide, but if you don't know the criteria, you may reverse it.

 
mytarmailS:

Try to take this model and create a sample of all entry points, throw all sorts of indicators, and train another model that will decide whether to enter the trade or not.

For this method I will not have enough indicators, because the initial sample of about 15 lines, and if you look later on the activation, there will be well over 5-6k entries, on this learn how not very much, and here my method allows you to continue training without changing the volume of the sample.

I think it's better to throw all sorts of indicators at once when training.

 
Aleksey Vyazmikin:

For this method I will not have enough indicators, because the initial sample of about 15k lines, and if you look later on the activation, there will be well 5-6k inputs, on this to learn how not very, but my method allows you to continue training without changing the volume of the sample.

I think it's better to throw all sorts of indicators during training.

I think it's better before and after.

 

I've been reading this thread for a long time and haven't joined the conversation.

To be honest, I don't understand, what market model do you analyze and teach NS? On the basis of candlesticks, candlesticks, indicators or their combinations? Or what is something secret and unknown to others?

The market model always is, was and will be like a shape of letters in the alphabet. The font may be different, but the recognizability is constant. We can have a wider conversation on this subject, if you are interested. After answering my question.

 
Uladzimir Izerski:

I've been reading this thread for a long time and haven't joined the conversation.

To be honest, I don't understand, what market model do you analyze and teach NS? On the basis of candlesticks, candlesticks, indicators or their combinations? Or what is something secret and unknown to others?

The market model always is, was and will be like a shape of letters in the alphabet. The font may be different, but the recognizability is constant. We can have a wider conversation on this subject, if you are interested. After answering my question.

By model here in 99% of cases we mean the NS itself.

Nice products, your approach to the market appeals to me.

 
mytarmailS:

By model here in 99% of cases they mean the NS itself.

Beautiful products, I like your approach to the market.

mytarmailS:

By the model here 99% of the time they mean the NS itself.

Nice products, I like your approach to the market.

Thanks for the compliment, but that's not the answer.

It is clear that you teach NS, not the rabbit)((

 
Uladzimir Izerski:

Thanks for the compliment, but that's not the answer.

It's clear that you're training the NS, not the rabbit.)((

Ahahaha))

Yes as not an answer, maybe you want to hear something else?


There are signs on which there is a training

There may be anything - indicators, candlesticks, time, log rules, other NS ... Whoever wants what he wants, that's what he wants .....

there is a target, it's what we want to teach

We train NS, sex of the trained NS is called a model ...

I suspect that by a model you mean something else, something like a mathematical description of the market, its laws, etc... But here a model means something else, the way I described above.

 
mytarmailS:

Ahahaha))

Yeah as not an answer, maybe you want to hear something else?


There are signs on which the training goes

There can be anything - indicators, candlesticks, time, etc., log rules, other NS ... Whoever wants what he wants, that's what he wants .....

there is a target, it's what we want to teach

We train NS, sex of trained NS is called a model ...

I think that by a model you mean something else, something like a mathematical description of the market, its laws, etc... but here by a model you mean something else, what I described above...

I understand your jargon. Signs are not a model, but a set of model components.

But before the NS can be trained, there has to be a set, a smart set that can be trained. After all, you know yourself that no matter how you teach a fool, the wolf has a longer tail.

 
Uladzimir Izerski:

Yeah, I understand your jargon. Signs are not a model, but a set of model components.

But before you teach NS, there must be a set, a smart set that can be taught. After all, you know that no matter how you teach a fool, the wolf has a longer tail.

Uh-huh right you know).

Well, as long as we teach the fool) ), you got it right...

signs or their sum should be connected with the target, NS simply finds these complex relations ... if you put data that have no relation to the target you get what you get - garbage

 
mytarmailS:

Mm-hmm. You got that right.)

Well while we're teaching the fool ) , you get it right...

signs or their sum should be connected with the target, NS simply finds these complex connections ... if we put data which are in no way connected with the target we get what we get you garbage

Right. That makes sense to me. But teaching the NS is the second step. Why do we have to torture the fool, and not pay attention to signs. After all, it is primary and what follows depends on the primary.