Machine learning in trading: theory, models, practice and algo-trading - page 2887

New comment
 
Uladzimir Izerski #:

Each bar has an internal structure. If the structures coincide, it may be some condition.

A bar can be considered as a complex structure.

The example shows 5 minute bars in 1 hour. The hours are sliced not at the beginning of the hour, but I think the point is clear, that there is a difference between looking at a bare hour bar or a structural bar.

there are no working patterns with a fixed number of bars on the market, as I wrote above (at least I have not found any).

And using patterns with a dynamic size is a challenge (I mean within the framework of MO), there are methods for these purposes, at least the same wave analysis, but then MO is not needed at all.

 
Valeriy Yastremskiy #:

I can't understand what is the difference between increments and absolute price difference in the window. Besides, you can train not only on increments, you can train on relative changes from absolute price, or on logarithmic changes from absolute price)))))

returns are the difference x[i] - x[i-1]

and sometimes you need x[ i] - x[i-1044 ]

 
I want to run several strategies on historical data. Please recommend ready-made solutions for modelling!
 
mytarmailS #:
1) in order to trade something, it must first be properly analysed, increments are not suitable for analysis, because knowledge of past prices is lost.

2) what is meant by analysis? According to Wikipedia, analysis is the splitting of a whole into parts for study. What prevents me from doing it visually?

3) if for example a working market pattern is a false double top breakout.
We have a complex sequence of events and not statsyonary in time, what is the use of comparing it with SB??? Will we get something adequate at the output?

4) I meant that when analysing the market one should always keep in mind that it is a sum of separate events, participants, actions, that is why it does not repeat itself, too many combinations....
I also meant that when analysing the market it is necessary to decompose it into participants, or their actions or something else related to them, and then analyse it, by the way, this will correspond to the concept of the word analysis.

5) I am not competent in this

1) Almost always the target variable is either an increment or something related to it. Incorporating increments into features is another matter. But very often simple transformations allow us to see the relationship of traits to increments. For example, the difference of averages can be written as a linear combination of increments, etc.

2) Personally, apophenia gets in my way. It's hard not to see something you really want to see. I'd rather have some way of measuring the significance of levels - the return to them, for example.

3) SB is quite good at drawing double tops. It is important to check if there is any difference from the SB associated with this particular pattern.

4) Well, we have singled out the state from the list of participants. One list of possibilities of its influence on the market is several pages, but how can you find out when and what it applies from this list? Well, yes, there is fundamental analysis, but it is also not a panacea and it is difficult to do.

 

Ideally, according to tsos, you should take ticks, filter them and downsample them, otherwise aliasing will appear

But forex is not Tsos, it's a different physics, right?

 
Игорь Егоров #:
I want to run some strategies on historical data. Please recommend ready-made solutions for modelling!

If you want to do everything yourself, there is a CodeBase section with many ready examples of Expert Advisors and indicators. You can use them as a basis for your research. In this case you can test strategies using MT4/MT5 tools. MT5 also has integration with the python programming language. You can quite easily upload the required historical data and work with it. Here is an example of the upload function






import  MetaTrader5 as  mt5

mt5.initialize(timeout=10000)
print(mt5.terminal_info())
print(mt5.version())


def get_data(symbol,  time_start,  time_stop,  count=0):
    name_stocs = ['time', 'open', 'close', 'tick_volume', 'spread', 'low', 'high']
    tf = mt5.TIMEFRAME_H1

    if  count == 0:
        dataset = pd.DataFrame(
            mt5.copy_rates_range(symbol,  tf,  time_start,  time_stop),
            columns=name_stocs).set_index('time')

        dataset.index = pd.to_datetime(dataset.index, unit='s')
        dataset = dataset.reset_index(drop=True)
    else:
        dataset = pd.DataFrame(
            mt5.copy_rates_from(symbol,  tf,  time_stop,  count),
            columns=name_stocs).set_index('time')

        dataset.index = pd.to_datetime(dataset.index, unit='s')

    dataset = dataset.reset_index(drop=True)

    return  dataset

To test your strategies you need a tester written in python. I have posted mine in this thread (you can search under All Posts in my profile).


If you don't want to bother, there is a Freelance section where you can make a trading robot/indicator for your strategy for your money.

 
Andrey Dik #:

there are no working patterns with a fixed number of bars on the market, as I wrote above (at least I have not found any).

And using patterns with a dynamic size is a challenge (I mean within the framework of MO), there are methods for these purposes, at least the same wave analysis, but then MO is not necessary at all.

MO should be supported and applied, but at other levels of understanding of conditions for application.

No standard MO model can give a ready result head-on. But there are workarounds to apply MO.

 

Pattern recognition, who's up for that kind of hassle?

Or through wavelets with downsampling.

 
Rorschach #:

Pattern recognition, who's up for that kind of hemmorhage?

Or through wavelets with downsampling.

Pattern recognition or market model recognition is the first brick.

It can be done with MQL tools, but with MO this method will be more advanced and progressive.

P.s.

We can look into the future more boldly.

 
Rorschach #:

Pattern recognition, who's up for that kind of hemmorhage?

Or through wavelets with downsampling.

This may have worked when most of the money was handled by visualisers. Now it's mostly matstat and MO. The fact that we have a predominance of retired visualisers here on the forum should not be confusing.

1...288028812882288328842885288628872888288928902891289228932894...3516
New comment
Reason: