Machine learning in trading: theory, models, practice and algo-trading - page 2427
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
Well, I here, since making a script to prepare the data, you still need to make a file listing the excluded columns, which include:
1. Columns with correlated predictors (by the way, how do you choose which column to drop, say 5 correlated predictors?)
2. Columns discarded from the first file-table, except for the column with the target.
Plus the column with the target label should be written into the file, preferably searched by the name of the column.
The structure of the file is as follows
I need to select the desired predictors in less time. To go through the predictors again is to increase the processing time by hundreds of times. My method is based on the logic that a good predictor (including one suitable for a particular learning method) will be demanded by the model at all sampling intervals, which excludes fitting to the sampling area.
In order to have it on all of them, we need to cross validate. And you only check by test or exam.
If you cross validate 10% of the sample, you need to train 10 times, not hundreds. And if you do 20%, then 5 times.
To have it for everyone, you have to cross validate. And you check only by test or exam.
If you cross validate 10% of the sample, you need to train 10 times, not hundreds. And if it's 20%, it's 5 times.
I break the sample into 8 sections and build 100 different models for each section, then analyze the models and see which predictors were in demand - which means that they were used to find a pattern, I average the estimated value, and then use the remaining predictors for training on the entire sample. The logic is that since the patterns from these predictors were found in an individual plot, the model will be able to generalize on these predictors to the entire sample evenly, rather than adjusting for sample areas as it usually does.
Your method involves building a model on a small part of the sample, and the models will be built differently in each area, as the best predictors will be selected that fit a particular training area, and given the fact that the sample is incomplete (representative), we can argue that this way only part of the available information is studied, which may or may not repeat in the future, my method will allow to learn more information about the market, with less overtraining. Besides, if in CatBoost not to fix the quantum table, then each time in general training will be on different predictors because of different constructions of quantum tables under the concrete site of sample.
Now the answer to the first question
I don't know what's wrong - it swears.
I don't understand what's wrong - it swears.
Run the function creation code and then the function itself
It worked, thank you.
Let it be your homework)...
I could only use concrete examples to figure it out. I would rather use MQL to solve this problem.
The most mystical forum thread now with telepathic communication.
So no one has admitted what kind of car they teach here and what
I also wonder why every scarecrow who sells his wretched averaging machine considers it his duty to promote himself in this thread, is it a dedication or what? ))