ForumSections

Machine learning in trading: theory, models, practice and algo-trading - page 1162

New comment
mytarmailS:

I wrote it, especially for those who are in the tank

Well, then publish for us tankers chart with a forecast for December, so that the prices, not like not seen, and really

Philipp Negreshniy:

If you do not see the forecast for December, you may leave it on the chart for us tankers.

So what's the point of making such a prediction? Why if you can separate the data into a trace and a test and check it at once, which I did?

In general, the essence of my message was not about predictions, but about the nature of reversals. Personally, I found this feature very interesting and decided to share it...


If you want a forecast, keep it, the reversal should happen on one of these arrows (in theory)

mytarmailS:

So what's the point of making such a prediction? Why if you can split the data into a trace and a test and check it at once, which I did?

In general, the essence of my message was not about predictions, but about the nature of reversals. Personally, I found this feature very interesting and decided to share it...


If you want to make a forecast here, you may use this one, the reversal should happen on one of these arrows (according to the idea)

it looks like a four-hour eurodollar, if so, then somewhere at the end of next week predict growth, thanks, we'll see...
Philipp Negreshniy:
this looks like a four-hour eurodollar, if so, then somewhere at the end of next week predict growth, thank you, we'll see...

it's a 5 minute eurodollar 8k treyn points and 8k forecast

mytarmailS:

this is a 5 minute eurodollar 8k points trein and 8k forecast

yes even 100K, write the essence of the forecast - how the price will change and when:)
Hi all. Who wants to try their TS on my data? Let's try again. Specify the minimum number of records for training. I remember called the numbers from 1000 or less?
Mihail Marchukajtes:
Hi all. Who wants to try his TS on my data? Let's try it again. Specify the minimum number of records for training. Remember called the numbers from 1000 or less?
TS is a trading strategy and how is it to try - to trade on this data?
Philipp Negreshniy:
How do I try to trade on this data?

Indeed, especially since at the moment it was only possible to save 200 records, which for many will be exceptionally small. So there's no point... let's pass...

Maxim Dmitrievsky:

How is your network Max? Share it, it's interesting

mytarmailS:

Have you ever tried this?

We take one of the most important components, the second or the third one , if it has a pronounced periodicity, we don't take the first one, because it usually has a pure trend.

I haven't tried it, I did everything in Matlab, because I had a lot of ready-made material, but my license for Matlab disappeared, and I only got it working again today... but I don't think I'm going to "poke around" on this subject any further

The problem is in periodicity, which does not and will not exist, any method - SSA, or Fourier, or MLP neural network - implies that the periodicity found on the input data will continue in the future, alas, there is no such thing with price series, and if it were possible to find periodicity, then the hundredth time, the standard indicators from MT supply would work better than any modern matrix)))


Maxim Dmitrievsky:

Ohoo... an article on how to analyze the date

For those who will ever get the hang of it :) I'm thinking of translating the application for the market, with some simple example like a watch that measures heart rate, like here

https://towardsdatascience.com/markov-chain-monte-carlo-in-python-44f7e609be98

It's not 100% accurate, but real-world data is never perfect, and we can still extract useful knowledge from noisy data with the right model!

I'll read it at night, if it's an article on how to convert input data (price series) for further analysis, then that's what I'm looking for

1...115511561157115811591160116111621163116411651166116711681169...3749
New comment