Machine learning in trading: theory, models, practice and algo-trading - page 1833
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
what pros??? they want to predict the market by reading tweets
You can always talk, as long as you have a tolerable interlocutor))). If you create something, it all starts with a problem statement and its understanding.
Algorithm to find events then repeat a chain of events and make a prediction based on an incomplete image, or what is wrong?
Or something else?
Let's go another way, write what you do not understand, I have twice drew a picture, I think I've explained everything, but I explain badly, I certainly do not have that
Let's go another way, write what you do not understand, I have twice drew a picture, I think I've explained everything, but I explain badly, I certainly do not give it
The algorithm to find events and then replicate the chain of events and an incomplete image make a prediction.
This is what I understood from the pictures, is this correct or not?
I did it yesterday, sometimes it works.
They want to predict the market by reading tweets.
They want to bolt on FA most likely through MO, but they need some logic there first, though maybe they'll find a correlation with social networks or marketplaces all of a sudden.
And from m-plays it is possible to screw the sales graph to the price of shares indirectly.
Algorithm to find events then repeats the chain of events and on the incomplete image make a prediction.
This is what I understood from the pictures, is this correct or not?
For example clustering the price into 10 clusters.
What is a cluster is a kind of pattern that repeats itself - a "pattern"
A cluster has a number v. days. case from 1 to 10
The price in the form of clusters will look like this
1113333555433377779991010103333222288888
we choose the combination at random e.g. 1593
look for it in the price
1113333555433377779991010103333222288888
this will be the resultant pattern
put all situations with this pattern together and look at the statistics of profitability
there may be hundreds of clusters, you may build them on any data, clusters may be replaced with log rules for example, it will be more clear
For example, we cluster the price into 10 clusters...
What is a cluster - it is a kind of image that repeats - "pattern"
The cluster has a number v. dn. case from 1 to 10
The price in the form of clusters will look like this
1113333555433377779991010103333222288888
we choose the combination at random, for example 1593
look for it in the price
1113333555433377779991010103333222288888
this will be the resultant pattern
put all situations with this pattern together and look at the statistics of profitability
there may be hundreds of clusters, you can build them on any data, you can replace clusters with log rules for example, it will be more clear
What data should we use? Don't bullshit with numbers, just clusters
with what data? it's not very good with increments
What data should be used to plot? Don't do this crap with numbers, just clusters.
What kind of data? It doesn't work very well on graphed data.
increments are the dumbest thing you can do with prices, then AMO doesn't even see a linear trend in the data
Let me draw it.
first we need a lot of prototypes (data preparation)
go by the sliding window on the data, in the sliding window there is always the last price
we don't need all prices because prices beyond 100 pips of the last price hardly influence the last price
That's why we cut out unnecessary data and leave only the green range in price
then
and we renormalizedin price vs.last price
then broke it into clusters
So you teach the model to see the same clusters in the in price range on different data, when you learned it you start looking for patterns
increments is the dumbest thing you can do with prices, then AMO does not see even a basic linear trend in the data
Let me draw a picture.
first you need a lot of prototypes (data preparation)
go by the sliding window on the data, in the sliding window there is always the last price
ingenious ))
it sees everything in increments, but the error is too big
We don't need all prices, because we don't need prices beyond 100 points of the last price to influence the last price.
That's why we cut out unnecessary things and leave only the green range in price
then
and we've renormalizedin price relative tolast price
and if there's a strong trend, you'll have almost no prices in the range
And always a different number of features