Machine learning in trading: theory, models, practice and algo-trading - page 1740

 
Maxim Dmitrievsky:

and color the clusters on the graph later, if you can... it's hard in python

to see the length of the clusters, how they alternate, etc.

I made three.

Is it the same for you?

 
Maxim Dmitrievsky:

No, I have a beauty, I showed you... I don't know why) something someone is doing wrong

traine

test


show me a fragment of the sample, which returnees you took and how

 
mytarmailS:

show me a fragment of the training sample, what returns you took and how

Take increments 5 and 25 for clustering

and then use the cumulative one. 1. On the Eurobucks hourly. Test and traine by year.

 
Maxim Dmitrievsky:

take increments 5 and 25 for clustering

and then use the cumulative one for increments. 1. On the eurobucks hourly. Test and trace by year.

Let's do it tomorrow, I have my eyes full

I feel dumb
 
mytarmailS:

Let's do it tomorrow, because my eyes are already popping out of my head.

I feel dumb.

There will be strictly one curve upwards, one horizontal and one downwards, there is no other

it's like a test case

 
Mihail Marchukajtes:

What we see, the more stable the period, the more round. That is, I can take wavelets, for example, and get the same picture.

In general, it does not work very well. The picture below has good periods (2 and 3), but they are not very smooth and make the circle fall apart.

It sayshere that cssa is an ssa built on a neural network prediction. That's what I wrote before, lag can be eliminated only by prediction. In usual ssa most likely instead of prediction they duplicate last known price, but in cssa they build prediction with neural network.
 
Maxim Dmitrievsky:

I have to go to sleep, I felt that I was dumb... The sad thing is that you are always predicting the past and adding up the past too, although on new data. For example if you found an "up trend cluster", the cluster itself will appear only after this trend has happened, post factum, the red dot is the moment when the cluster appears

and then you put it all together and you get a nice picture of an up trend.

but it's post factum, you know?

The cluster will say "trend" when the trend has already happened.

 
mytarmailS:

Max, it's bullshit with those clusters, it's good that I went to bed, I felt that I was dumb... The sad thing is that you are always predicting the past and adding up the past too, although on new data. For example, if you found an "up trend cluster", the cluster itself will appear only after this trend has happened, post factum, the red dot is the moment when the cluster appears.

and then you put it all together and you get a nice picture of an up trend.

but it's post factum, you know?

The cluster will say "trend" when the trend has already happened.

they should be relatively long and rarely change

 
Maxim Dmitrievsky:

they should be relatively long and rarely change

Okay, but you'll always be behind on the window size for the cluster

 
mytarmailS:

OK, but you will always be behind on the size of the window for the cluster

I don't really know, I'm too lazy to think about it... or should I?