Machine learning in trading: theory, models, practice and algo-trading - page 3462
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
I think it's easier to add a few columns for MO: day of the week, hour, minute. Maybe also month, day of month and/or year.
I think it's easier to add a few columns for MO: day of the week, hour, minute. Maybe also month, day of month and/or year.
Or he means whether the MO can understand from the signs without time what is the current time in the observation.....
I don't know how it turns people's heads that they go from converting BP, the original message, to time.
Okay.
Returns or their absolute values (volatility) contain the time of day, because each time period has a different volatility. To do this, we need to choose a corresponding lag of increments around 20-25 for the hourly TF.
Naturally. Naturally, the trading time period is different from the usual one.
It is elementary.I think it's easier to add some columns for MO: day of week, hour, minute. Maybe also month, day of month and/or year.
Perhaps instead of a measure of time it would be more reasonable to add statistical volatility (deviations).
For example, instead of the column "hour" - typical range for this hour, for the historical period. MO has a memory like a goldfish, and this would be instead of it
It may be more reasonable to add statistical volatility (deviations) instead of a measure of time.
For example, instead of the column "hour" - typical range for this hour, for the historical period. MO has a memory like a goldfish, and this would be instead of it
For convenience reasons and because MO works better with continuous data.
But the topic was different, how it always happens :)
if you regard time as making a choice for the solution instead of the network then yes in that sense its "a step in" and you are obscuring data for the machine.
But you can think of it as a convolution
Today I will try to train a model for 4 million features, that is, I will try to replace the usual pattern as a filter with a model as a filter.
This is a continuation of this topic.
#34553
But first, lunch)))
A cool terminal, replacing the standard one, with inbuilt AI assistant. You can write code directly in the terminal, for example in python.
You can connect some text editor like nvim, as far as I understand, it will be a full-fledged lightweight IDE. I like these.
It's free at the moment.
Today I will try to train a model for 4 million features, that is, I will try to replace the usual pattern as a filter with a model as a filter.
This is a continuation of this topic.
#34553
But first, lunch)))
4 million columns? And rows? If at least 100k, that's probably months of calculations.....
4 million columns? What about rows? If at least 100 thousand, that's probably months of calculations.....
rows +-200
is a filter of one pattern, not an attempt to explain the whole market.
And 4 million is not enough, at this stage of my understanding of the problem I need 20 million, but I don't know how to cope even with 4 million, it's really a lot...
And the signs are already compressed 50-100 times from the usual representation...