Machine learning in trading: theory, models, practice and algo-trading - page 293
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
We are not creating a hadron collider here, you don't have to be a mathematical scientist to be able to trade.
So what are you doing here then?
Go to a branch where they study mash-ups or stochastics optimizing, everything is simple and clear there... You're not the first or second such that comes into this thread, you have zero knowledge, nothing to say, so I'll propagandize that neural networks are evil and so on ... And here the opposite I have not seen, that someone propagandize networks to those who study mashki ... Why are you all climbing here? No one invited you and no one is climbing your way either.
Good afternoon! Can you tell me, for an LSTM network, in what form the input parameters are fed. As I understand there is a system of switching "gates".
As far as I know, feed a vector (array) of numbers into the input, count the result, then feed the second array of numbers, count the result, etc..
It turns out that when calculating the result, the incroute passes through all neurons, and these neurons remain in a certain altered state, as if they remember the previous incroute.
I do not know anything about neural networks, I see people here who experiment with them, but probably before that they worked with systems based on indicators and other pattern visualizers.
The question is this: Has anyone tried to use a neural network to test the effectiveness of TS with a lot of variables on the results of the TS - i.e. the results of the report to identify the most significant variables and the effectiveness of their impact on the TS?
So what are you doing here then?
Go to a branch where they study mash-ups or stochastics optimize, everything there is simple and clear... You're not the first or second such that comes into this branch, knowledge zero, have nothing to say, so I'll propagandize that neural networks are evil and so on ... And here the opposite I have not seen, that someone propagandize networks to those who study mashki ... What are you all doing here? No one asked you to come here and no one is bothering you either.
I'm just disgusted that the "scientists" here are confusing people, demotivate them with their academic degrees and the number of educations, tempting participants to spend a lot of time on 100% unpromising trading systems. The market does not need neural networks, they exist since the 80's and since that time they are used in the markets by mathematicians of the highest level, no one particularly excelled in their use in the market, everyone agrees that simple systems based on indicators are much more effective, certainly not on the usual wrenches and stochastics, but for example onJMA with good money management.
Neural networks and complex models are fed to the "meat" puppets to distract from trading for many years and to disappoint in their intellectual abilities. In fact the cool hedge funds use JMA and I think mangindale does too, that's why they have such nice yield curves. The important thing is to find a trading system that meets your needs and strictly follow the rules of your TS, all the rest is evil.
1) The idea was like that, but I have not tried it, you need to define the characteristics of what to send to the network, and that there is a result (target), you can banal by equity, if it grows, then good, if it falls, then bad, but there are nuances
2) Many parameters is bad.
It seems to me there should be two phases (layers - as they call it) - the first one is looking for correlation of indicators - the highest correlation, and duplicate correlations are eliminated, and the second layer takes into account remaining indicators and adds EA variables - and we look at their influence on selected indicators. We strive for more profits, less losses, less drawdown and more profits per trade - at that, we give priority to the greater number of trades.
For the second day I sit and do such a thing with my hands, in TS there are 3 variables - and so everything is not unambiguous :)
By the way, maybe someone here knows, I have independent series of numbers (indicators of optimization results), which can be divided into subgroups, so I do normalization of indicators in each subgroup - the size of subgroups is the same, but if I do a correlation before and after normalization, then sometimes there is no correlation where it was before. I check the correlation on the whole number series. What does this effect mean? There is an assumption that the correlation was false and normalization revealed it. Yes, I do the normalization as a percentage of the maximum index in the group - the purpose is to compare numerical series with each other.
I'm just disgusted that the "scientists" here are confusing people, demotivate them with their academic degrees and number of educations, tempting participants to spend a lot of time on 100% unpromising trading systems. The market does not need neural networks, they exist since the 80's and since that time they are used in the markets by mathematicians of the highest level, no one particularly excelled in their use in the market, everyone agrees that simple systems based on indicators are much more effective, certainly not on the usual wires and stochastics, but for example onJMA with good money management.
Neural networks and complex models are fed to the "meat" puppets to distract from trading for many years and to disappoint in their intellectual abilities. In fact the cool hedge funds use JMA and I think margindale does too, that's why they have such nice yield curves. The important thing is to find a trading system that meets your needs and strictly follow the rules of your TS, all the rest is evil.
Forgive me, but you don't seem to understand where you're going and what the market is, I understand you would write a procluster analysis, delta and all, that really moves the market and is the reason for it. But you referred to JMA, which was developed in 2006 or so, adopted by people, taken apart by people on bones and understanding that JMA is slightly better than a MMA, literally just a little, that may result in profit some 10 years later with a risky MM. I don't think that cool Hedge Funds use JMA, most likely SOT reports, volume analysis, option levels, but not your JMA, and if you rely on indicators, then I have nothing to talk to you about, because you really don't understand what the market is and who its participants are.......
Friends, good time! Please advise how to do it correctly in Python.
Для примера построим график продаж видео игр в различных странах в зависимости от года.
Для начала отфильтруем только нужные нам столбцы, затем посчитаем суммарные продажи по годам и у получившегося dataframe вызовем функцию plot без параметров.
sales_df = df[[x for x in df.columns if 'Sales' in x] + ['Year_of_Release']]
sales_df.groupby('Year_of_Release').sum().plot()
The thing is that the name of the columns of numbers, how to correctly specify the numbers - the syntax?
sales_df.groupby(2).sum().plot()
It gives out an error
TypeError Traceback (most recent call last)
<ipython-input-75-494 b1527114a> in <module>()
----> 1 sales_df = df[[x for x in df.columns if 1 in x] + [2]]
TypeError: argument of type 'numpy.int64' is not iterable
What if it's in quotes? I know Python on a hello world level, but it might work.
sales_df.groupby('2').sum().plot()
What if it's in quotes? I know Python at a hello world level, but maybe it'll work.
sales_df.groupby('2').sum().plot()
I've already tried everything.