Machine learning in trading: theory, models, practice and algo-trading - page 2284

[Evgeny Dyuka]

Renat Fatkhullin:

Now you offer super-duper-duper new integration with python. So I'm sitting here thinking, why the hell should I get into it?
At what point would I need it?
You seem to be losing your sense of the market and ceasing to understand the customer.

[Forester]

Renat Fatkhullin:

Note to MQ.

In MO most of them use data of OHLC bars for training. It makes no sense to execute real ticks because they take considerably longer. For example, I use the tester by opening prices.

It would be desirable to have the second candle with OHLC prices at Ask instead of the minimal spread for Bid bars.

Thus, it would be possible to estimate EAs using real data without running real ticks.

For example, it is unlikely that Ask was equal to (Bid - minimum spread) at High Bid. The spread is different and may be even several times larger than the minimum one.

And the High Ask could be not at the time of the High Bid, but when the Bid has already decreased slightly.

The second candle by Ask would give the possibility to correctly estimate the trade. For example, if the tester works with the TP/SL or with trailing stops, taking into account the minimum spread in the candle, it will say that the TP for a buy trade triggered. And in reality the Ask was lower, because the spread was not minimal at that moment and the TP might not trigger. I.e., the tester will show different results from the real trade by the opening prices and by OHLC.

In fact, all non-MO EAs (those that work with TP/SL and trailing stops) would have been tested more correctly at opening prices and by OHLC if the OHLC Ask was known.

[Aleksey Nikolayev]

Evgeny Dyuka:

Now you are offering a super-duper-duper new integration with python. So I'm sitting here thinking, "Why the hell should I get into this?
At what point would I need it?
You seem to be losing your sense of the market and ceasing to understand the customer.

You are making the wrong assumption that MQ's clients are traders. Most of their customers (i.e. those who bring in real money) are retail forex DCs. But they want to diversify the set of clients - hence the support of python as an attempt to attract large investment funds.

[Evgeny Dyuka]

Renat Fatkhullin:

By the way, the idea of taking a bite out of tradingview pie is very realistic with your potential of programmers.
You have to go in from crypto side and develop this direction. I'm moving the visualization of my neuronet to a browser because showing its work in MQL5 terminal is a real pain - nobody uses that old-school terminal and does not want to, you can hardly make a client install it. And even if you do, they will ask you a lot of questions later.

[Aleksey Mavrin]

Renat Fatkhullin:
Can you share the information:

1) Do you use the python library of MT5?

2) Do you use it outside or inside MT5?

3) What functions does the library lack? Access to indicators?

We are preparing an upgrade of MQL5 adding fast matrix operations. This will allow to perform massive calculations normally.

We will also develop connectors to analytical packages and implement the standard WinML integration.

1) Not yet, but definitely will have to. While for ML I use solutions written in MQL by the habit of working "all in one place".

2,3) Haven't figured out how to use inside, I think mqh-wrapper interfaces for popular ML Python libraries would be in demand.

Will there be matrix operations with GPU computation capability?

[Evgeny Dyuka]

Renat Fatkhullin:

4) has always been available

This is your answer to the question about the possibility to post an Expert Advisor with webrequest in your marketplace.
What do I do with this answer? I wrote an EA for sale, tried to put it up, they turned me down. It took me a shitload of time and effort and that was six months ago. Maybe I'm stupid, but I'm not going to look twice to find out where I should have seen a button and done something right.

This is an example of losing a client. I understand that "when the winged horse Hay-Fay rushes down the mountain, he has no time for toads sitting by the road, but so soon MQl will simply dissolve into history.

[mytarmailS]

Evgeny Dyuka:

By the way, the idea of taking a piece of the tradingview pie is very realistic with your programming potential.
You have to go in from the crypto side and develop this direction. I'm moving the visualization of my neuronet to a browser because showing its work in MQL5 terminal is a real pain - nobody uses that old-school terminal and does not want to, you can hardly make a client install it. And if you do, they will ask you a lot of questions.

You're right, you should have done so in the first place.

Simple, quick, understandable, clear, all the controls you have...

[Evgeny Dyuka]

mytarmailS:

That's right, you should have done that right away...

simple, fast, clear, visual, all the controls you have...

node.js has to be mastered, and that's a pain

[Rorschach]

Evgeny Dyuka:
node.js is a pain to learn

Have you ever used Brython? It's python for the browser.

[Renat Fatkhullin]

Aleksey Vyazmikin:

Please make the OHLC synchronization mode correct, so that at least standard indicators do not have glitches when requesting data from the upper TF.

Otherwise there is no point in python to get data from indicators, because training on all ticks is suicidal.

Another annoying thing is slow speed of reading/writing files (csv/txt) in MT5.

If we are speaking about synchronization of two MqlRate/MqlTick arrays by date with completion of missing values, then most likely it will be done by the standard function. This is a common situation when comparing/correlating history of different symbols.

If we are talking about synchronization of MqlRate and double arrays, then there is no synchronization point in the form of date.

Specify exactly what and how you mean.

You need a code to look at the speed.