Machine learning in trading: theory, models, practice and algo-trading - page 1876

New comment
 
Maxim Dmitrievsky:

and the days can be different for these hours, no?

I had the same thing, because of this there were inconsistencies in the charts, and you do not have it for some reason

They can't!!! Because I'm taking the next one from the beginning of the 5 o'clock. I don't cycle up to the sixth hour like you do, you know?

You can jump a few days, I can't.
 
mytarmailS:

They can't!!! because from the beginning of the 5 o'clock I just take the next one. 23 indexes, and I don't cycle to the sixth hour like you, you know?

You can jump a few days even, I don't

Anyway... there is a gap in the dataset (at least one hour). For some reason you can't find it on the graphs )) I dunno

for other hours there are an unequal number of 5-minutes, then there are also gaps. And when everything is the same, then there are no gaps in the graphs

In short, I'm tired of explaining.

If you shifted by at least one hour, there will be a cascade of gaps in prices, like this... after that hour everything shifted

My code works this way, it means it's sensitive. But all fixed (hooray)


 
Maxim Dmitrievsky:

fixed (hooray)

super

 
mytarmailS:

super

but not the code itself, but added the missing indexes to the dataset via reindex, and the gaps disappeared

prices = prices.reindex(pd.date_range(startdate, enddate, freq='5min'))
prices = prices.fillna(method='backfill')
 
Maxim Dmitrievsky:

but not the code itself, but added missing indexes to dataset via reindex, and the gaps disappeared

let's see some trade results of the bot, otherwise it's not interesting at all

 
mytarmailS:

Let's get the bot trade already, because so far it's not interesting at all.

Let's just stop and do nothing further, we don't need these bots

i confess i just like writing python code
 
Maxim Dmitrievsky:

Let's just stop and do nothing further, we don't need these bots

i confess, i just like to write python code

))))) will do)

 
The problem is with the skips. I understand that the 5minutes do not beat the clock, the opening of the lower ff is not tick to tick with the older, somewhere on the forum was noticed. But 5 minutes without ticks is too much. If only time in the middle of 5 minutes, and then there is a shift. But still 2.5 minutes. something is wrong.
 
Maxim Dmitrievsky:

Anyway, check it out. I'm going to the mountains for a week.

Here are the clusters for 1-2 hours each day from 2010 to the present day.

here's the tree of life in the attachment.

You can trade flat or directional clusters to try. For now for these clocks. As soon as I get there, I will do it myself.

In the function, pass 12 5-minutes of the last 5th hour (close), normalized to the 1st five minutes of the hour, the function returns the number of the cluster. At the 6th hour, we trade. For the 6th hour, we also normalize by the 1st 5-minute of the 5th hour.

Tree classification errors:

>>> print(train_score, " ", tst_score)

1.0 0.8327272727272728

are the brackets filled in?

 
Valeriy Yastremskiy:

are the brackets checked?

Yes, it's working.

try to compile in the editor

corrected the time in the message text... it's 1-2 instead of 5-6

Eurobucks, respectively
1...186918701871187218731874187518761877187818791880188118821883...3526
New comment
Reason: