Machine learning in trading: theory, models, practice and algo-trading - page 68

 
Mihail Marchukajtes:
And anyway, what is -1 in the output. Black and white says 0 or 1 minus 1, the predictor treats it as zero. So what to do?
No one canceled the selection of inputs. This is one.

To recode -1 to 0 is a task for the school. That's two.

Three classes is a strong move up, down or a flat.
 
Mihail Marchukajtes:
Send me a file with the data, I will apply the model to you and you will see how it will work....
How will you send me the model when I don't use Reshetov classifier, and my data have 60 -560 predictors and 25000 observations, probably not enough for you to learn a month
 
Alexey Burnakov:
This idea is fresh but it is not clear. Is it working or not.

It seems to me that it works.

If you take two filters as a support, so that they overlap. All filters overdraw to the right, but if you take 10 bar instead of 1 bar, it's okay. We get a very good support.

On historical data we look ahead from crossovers and mark crossovers +1 if we get +100 pips in the future and vice versa. We add predictors and select those that have predictive power for that target variable.

That's it.

 
Alexey Burnakov:
Nobody cancelled the selection of inputs. This is one.

To recode -1 to 0 is a task for the school. That's two.

Three classes is a strong movement up, down or a flat.
All right, I will recode -1 to 0 and let's train the model for growth and then for decline when 1 will be recoded to 0 and we will have two models. But let's try to make one first. Taking into account that there are a lot of entries I think the training will last a long time, but I'll try. The most recent data at the bottom?
 
SanSanych Fomenko:

On historical data, look ahead from crossovers and mark crossovers +1 if you get +100 pips in the future and vice versa. We add predictors and select those with predictive power for that target variable.

That's it.

I did the same thing with simulated stop loss, but it didn't work... I think I need something else.
 
Mihail Marchukajtes:
All right, I'll recode -1 to 0 and let's train the model for growth for the time being, then for decline when we recode 1 to 0, we'll have two models. But let's try to make one first. Taking into account that there are a lot of entries I think the training will last a long time, but I'll try. The most recent data at the bottom?
It's a more complicated structure. There are five pairs. In each sequentially from the old to the new dates.
 
Alexey Burnakov:
There's a more complicated structure. There are 5 pairs. In each of them sequentially from the old dates to the new ones.

I don't know, the data is too much I will leave the computer until morning, but if nothing comes out in the morning.... Anyway, we'll see how it goes. You can then apply the model and see what happens out of sample. And in the studio as a graza for general review????

The thing is that sometimes the predictor needs to be trained several times to get acceptable data, but I will do it only once. ok. what will work out will work out....

 
Mihail Marchukajtes:

I don't know, the data is too much I will leave the computer until morning, but if nothing comes out in the morning.... Anyway, we'll see how it goes. You can then apply the model and see what happens out of sample. And in the studio as a graza for general review????

The thing is that sometimes the predictor needs to be trained several times to get acceptable data, but I will do it only once. ok. what will work out will work out....

I'll give you the out of sample as well. I don't use Yuri's software...
 
Alexey Burnakov:
I'll give you out of sample. I don't use Yuri's software...
How do you check if the model works? I thought I will send you the model, you will put it in your EA, run it on the history and get a curve of balances and give it to the public. how so.... Tell me right away will you be able to do it? otherwise I may be training for nothing....
 
Mihail Marchukajtes:
How do you check if the model works? I thought I'd send you the model, you'd put it into your EA, run it on history and obtain the balancing curve and display it for all to see. how so.... Tell me right away will you be able to do it? otherwise I may be training for nothing....
Nope. I do not use his self-written software. I am using R and the Expert Advisor for R.

Don't tell me that you're deflated) you can check outside the sample yourself using my data.