Machine learning in trading: theory, models, practice and algo-trading - page 3660

[mytarmailS]

Maxim Dmitrievsky #:

Well the same kind of thing works for k-mins )

Kmins clusters randomly.

[Aleksei Kuznetsov]

Maxim Dmitrievsky #:

Well the same kind of thing works for k-mins )

mytarmailS #: Predictors you pass through yumap then into hdscan then you look for such clusters in the occurrences of which the odin class label is often found....

The classifier will work fine on new data.

Would someone make a signal. to see what's going to happen on the oscillator.

[mytarmailS]

Forester #:

Would someone make a signal... just to see what's going to happen on the O.O.C.

I've done a lot of tests in the past and experiments, the bunch I mentioned really works on OOS.

But you'll have to take my word for it.

[[Deleted]]

mytarmailS #:
Kmins clusters out of the box.

Not always, depends on the data and the number of clusters. The speed allows you to go through the variants.

[[Deleted]]

Forester #:

Would someone make a signal... just to see what's going to happen on the O.O.C.

What it looks like on the OOS is how it works.

A couple of active accounts. There are more of them, different models.

Models are generated at a rate of one in 10 seconds :) Not hft, for normal deposit management more.

This is from reals charts.

[[Deleted]]

Dropped off the size improvement guy, there were a few options. I don't know who tried it.

Could improve hopeless partitioning by clustering.

[Aleksei Kuznetsov]

Maxim Dmitrievsky #: Models are generated at a rate of one in 10 seconds :) Not hft, to manage normal deposits more.

This is from the reals of the graphics.

Looks encouraging)
On the first screen the income is 130 on 230 trades. And on average in pips per trade how much?

[[Deleted]]

Forester #:

Looks encouraging)
On the first screen the income is 130 on 230 trades. What is the average in pips per trade?

I haven't counted, I'll have to go to the vps again. I'll look tomorrow.

The models are different, so I don't emphasise it.

[[Deleted]]

At the same time the algorithm can work on a single MA, for example with a period of 30 :)

>>>  hyper_params['periods']
[30]

This is to the question of feature search. Or rather, a dilemma that doesn't really exist. Just fewer signs will explain less variance.

[СанСаныч Фоменко]

mytarmailS #:

Recipe for classification GUARANTEED without retraining

Take the class labels

Predictors you pass through unmap then into hdscan then look for such clusters in which the same class label is often found....

That's it... :)

The classifier will work fine on new data

The highlighted part is the most important.

It's been my foundation for years.

BUT.

It's no use at all.

OOV and OOS are all fine. And if you run it in a tester or simulate a tester in R by moving the window one step forward, the result has nothing to do with OOB and OOS.

Sad. this is exactly the result that does not depend on the preparation of predictors.