Machine learning in trading: theory, models, practice and algo-trading - page 384

[Алёша]

Maxim Dmitrievsky:

I did not even try 8h, went straight into the full free version, but it needs an account

Yeah, I found it, I logged in, I'm playing...

And how do I specify where the features are and where the targeting is?

[Maxim Dmitrievsky]

Aliosha:

Yeah, I found it, I logged in, I'm playing...

And how to specify where the chips and where the targeting?

The Train model module specifies the targets. I recommend to make a training example first, it's very easy to understand what to connect to.

https://docs.microsoft.com/ru-ru/azure/machine-learning/machine-learning-create-experiment

[Алёша]

Yes, it's nice...

If the result could also be exported to C++ code, compiled into a dll at least, it would be worth it!

[Maxim Dmitrievsky]

Alyosha:

Yes, it's nice...

If only the result could be exported to C++ code, at least to a dll compiled, it would be worth it!

Yes, it's only disappointing, that trained model can be handled only via web api... but nothing prevents to write everything in R, teach and develop it there and then ready to use in your PC. But in general the web api is also cool

[Алёша]

Well, not bad, as for a web application...

[Maxim Dmitrievsky]

Alesha:

Well, not bad, as for a web application...

Threshold slider to the left, it will be more accurate :) This app is in the azure cloud with a lot of processing power. I haven't done any big projects yet, but it should read very fast. If it's not fast enough, there is another one from Microsoft: https://www.microsoft.com/en-us/cognitive-toolkit/

I'll have to check it, train a grid with different parameters, chosen via GA and see how fast the studio copes with it

Performance comparison with TensorFlow and MxNet https://www.microsoft.com/en-us/cognitive-toolkit/blog/2017/06/microsofts-high-performance-open-source-deep-learning-toolkit-now-generally-available/

[Алёша]

Maxim Dmitrievsky:

Threshold slider to the left, it will be more accurate :) This application is in the azure cloud with a lot of processing power, I haven't done any big projects yet, but the idea is that it should read very quickly. If it's not fast enough, here is another one from Microsoft: https://www.microsoft.com/en-us/cognitive-toolkit/

I'll have to check, do a grid training with different parameters selected via GA and see how quickly the studio copes

Performance comparison with TensorFlow and MxNet https://www.microsoft.com/en-us/cognitive-toolkit/blog/2017/06/microsofts-high-performance-open-source-deep-learning-toolkit-now-generally-available/

It does not work more accurately.

[Maxim Dmitrievsky]

Aliosha:

It doesn't come out more accurate.

Is it necessary to do 2-class trees in this example? And what accuracy should it be?

I don't know anything about numerai yet, I'll read it)

[Алёша]

Maxim Dmitrievsky:

is it necessary to do exactly 2-class trees in this example? and what accuracy should be?

The more precision the better, which is not essential

[Maxim Dmitrievsky]

Aliosha:
Accuracy the more the better, what to use is not essential

This is what they give tasks? I understand it is the hedge fund, if you register, what will it give?

I will run with different models, so far the same as you got 0.5