Machine learning in trading: theory, models, practice and algo-trading - page 1118
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
As usual - there's nothing to talk about)))
Come on. How much data do you need to estimate them?
I do not need anything. Give me at least 2-3K observations.
Date, Raw, Processed, Target.
Unfortunately, it is impossible to find such a number for the entire period of work. And then what??? The model trained on them will work for how long??? Forever???
I do not need anything. itslek give, at least 2-3K observations.
Date, Raw, Processed, Target.
He has AI, this one will learn by 50. Your MO is just weak.
He has AI, this one will learn by 50. Your AI is just weak.
Exactly. I do not need the NS to work for a year after such training. If it works well at least the same 50 points, which will be 100% of the period of training this will be considered a success. And what's the point of cramming thousands of lines into it, filling her head with garbage and unnecessary data????
The funny thing is that Mishani's optimizer reconstructs dependencies well over small samples, which is actually his specialty in fact. That's what the book says. And he vaguely captures it by what replaces his brain.
The other thing is that you have to test on a large test plot all the same
And what is the point of cramming thousands of lines into it, filling its head with garbage and unnecessary data????
That's when the NS will classify something. At least it will try to generalize something, if it's possible at all.
That's when the NS will classify something. At least it will try to generalize something, if it is possible at all.
Everything is true if the area is finite and static, but in our case it is infinite and non-stationary, so increasing the sample leads to a decrease in the quality of training, as a consequence of which the model works poorly on new data.
To make profit in the market, all other conditions being equal, the percentage of profitable trades must be more than 75% and not less. This is with the condition of equal profit and loss. You have trained the network on 1000 data and the learning result is 60% as an example. What is the point of putting such a model in your work if it is trained poorly???? I'm sure you can't get a good result on a large area. I'm talking about generalized model and not retrained model... IMHO
Trend = 100k lines. On the remaining (test) you apply the model.
The metric is logloss. Show me the results. Trend =... test =...
I took the first 1-- lines from your file and ran the training. If the result will be higher than mine by 40 instances, I will consider your data better than mine. Let's see now...