Machine learning in trading: theory, models, practice and algo-trading - page 55
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
I'll add an intrigue -- you don't have to file a change in the rate of the traded item.
It's like dragging yourself by your hair out of a swamp.
Look for other sources of data as well.
May Profit be with you!
:)
If that's what you mean? Well, you've discovered America. I'd be happy to include more information, but apart from the price history, it's hard to get anything else in such an amount.
It's not true about pulling the price by the hair. And it is enough to cover the costs and be in the black. It is confirmed in practice, though on the history. And those who can't take it out are dancing with tambourine. I have some experience in the glass but I'm not sure yet if I can do something with them.
***Tired.
Will there be a State?
I suggest reading some good notes on building trading systems from the position of a quant (a senior quant in a large investment fund). The ideas seem reasonable to me.
...What do you agree or disagree with? What would you like to study better?
Well at least with strict separation into training and test sample by dates instead of preliminary random mixing of examples with equal distribution in the general sample and then dividing it into parts. After all, it may turn out that one part of the sample will contain mostly vertical trends, and the second part will contain lateral trends. If we do random mixing, the probability of gathering of similar patterns in different parts of the sample decreases.
By the way, this disadvantage is also present in the built-in MetaTrader strategy tester, i.e. it divides the training sample and the forward test strictly by dates. Therefore, a change of market trends close to the dividing line can lead to a deliberate overtraining.
Boy, get out of here.
In particular, I liked this one:
So, you strictly separate in-sample and out-of-sample; - strictly separate training set and validation set
you blind yourself to date ranges; - separation of data exactly by dates (before X day - training, after - validation)
You use Monte Carlo to avoid starting-point biases.
and you try various robustness tricks. - Apply robustness techniques.
What else do you do to ensure that you're not fooling yourself?
Everything sounds right (but I haven't tried everything myself), except the first one. It seems to me that if you use strictly one training dataset, and select predictors for it, then you get super-fitting of these very predictors, which may not be reliable in other time intervals. I now create a new training and validation sample each time I select predictors before training the model. It's better to re-create the samples 3 times and train the model and take average accuracy than to always use the same training and validation samples.
I like the one about protein neural networks:) . Sometimes due to an unfortunate topology they indeed produce very inadequate results.
***Tired.
Will there be a state?
No. You have to be smarter to understand it long ago. There is a development here.
Everything sounds right (but I haven't tried everything myself), except the first one. It seems to me that if you use strictly one training dataset, and select predictors for it, you get overfitting of these predictors, which probably will not be reliable in other time intervals. I now create a new training and validation sample each time I select predictors before training the model. It's better to re-create the samples 3 times and train the model and take average accuracy than to always use the same training and validation samples.
I like the one about protein neural networks :) . Sometimes due to unfortunate topology they do produce very inadequate results.
Yeah. I would really like to see their results. I have something to show.
And you?
However, the question seems to be rhetorical.
Yeah. Can you tell me how many years of development is going on, and where can I see its results? :)