Machine learning in trading: theory, models, practice and algo-trading - page 3415

[Aleksei Kuznetsov]

Renat Akhtyamov #:

mu is the middle of a segment, a cluster in this case, I take it.

If it were a circle, the formula would work.

I think these are the coordinates of the centres. If you read the text more carefully...
Not a circle, a hypercube.

[[Deleted]]

Aleksey Vyazmikin #:

As I understand it, you need mu. It is different for each predictor, hence the vector/array.

Hierarchical would be more suitable for you, if you need to use it like leaves

[[Deleted]]

mytarmailS #:
What signs would you include in the model if you want to predict whether the TS will earn on the new data for the next n counts?

The same as the main one. It is equal to trade/not to trade, no?

[mytarmailS]

Maxim Dmitrievsky #:

We can use the same ones as the main one. It's the same as trade/no trade, right?

No, it's a different angle on the signs, it's not about stochastics.

You train a model to trade, what have you got?

You got a model, you got a capital curve.

the signs on the trade.

• you can estimate the shape of the curve,
• you can calculate the characteristics of the trades,
• the number of trades,
• the average duration of a trade over time,
• sharpness ,
• drawdowns.
• MORE WHAT?

signs on the model

• modelling error
• number of signs selected by the model
• var importence
• some other characteristics
• MORE WHAT?

Another third class of features.

• make predictions on the capital curve
• confidence intervals
• MORE WHAT?

And then we put it all together and try to predict whether this TS will make money on the test.

[[Deleted]]

mytarmailS #:

No, it's a different angle on the signs, it's not about stochastics.

You've trained the model to trade, what have you got?

I got a model, I got a capital curve.

signs on the trade

• you can estimate the shape of the curve,
• you can calculate the characteristics of the trades,
• number of trades,
• the average duration of a trade over time,
• sharpness ,
• drawdowns.
• MORE WHAT?

signs on the model

• modelling error
• number of signs selected by the model
• var importence
• some other characteristics
• MORE WHAT?

There's a third class of features.

• make predictions on the capital curve
• confidence intervals
• MORE WHAT?

And then we put it all together and try to predict whether this TS will make money on the test.

I can't answer that.

[mytarmailS]

Maxim Dmitrievsky #:
I can't answer that.

You didn't even think it was possible to approach it that way, did you? Isn't that right?

[[Deleted]]

mytarmailS #:
You didn't even think you could come at it that way, did you?

What's the target audience?

[mytarmailS]

Maxim Dmitrievsky #:

What's the target

Will you earn money from the test yes/no

probability by binary classification.

Either the slope of the captal curve on the test by regression or there FV or Sharpe or how much the figure will earn on the test.

Or better together two models and classification and Regression.

Then it is possible to generate 1000 TS, then selected 20 best ones by probability that they will work on the test, then selected n best ones by regression by Sharpe regression.

Cool

This is a serious experiment.

[Aleksey Vyazmikin]

mytarmailS #:
Will ts make money on the test yes/no

probability by binary classification.

Either the slope of the captal curve on the test by regression or there FV or Sharpe or how much will earn on the test figure.

Or better together two models and classification and Regression.

Then it is possible to generate 1000 TS, then selected 20 best ones by probability that they will work on the test, then selected n best ones by regression by Sharpe regression.

Cool

This is a serious experiment.

It's a dabble. Until you can't detect the probability shift in a single sheet - models will be casting.

And to work with a sheet or a quantum segment - you need quite a lot of responses on histories, and this is not available, and without it there is not enough stat data... so the models will be questionable...

[Aleksey Vyazmikin]

Maxim Dmitrievsky #:
Hierarchical would work better for you, if you want a leaf-like layout.

It is quite possible - you should try it.

However, as far as I remember, it requires a lot of memory for intermediate calculations, and it's not clear how to transfer it to MQL5, and whether it can be used on new data....