Machine learning in trading: theory, models, practice and algo-trading - page 2409
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
An interesting observation (strange that no one even thought of it)
If you train the model in a small window of 100 on non-normalized prices, then the model is very good at predicting new data if they (prices) are in the same range as in the tray...
Past prices are important for the model, i.e. levels...
The model fixes not only trailing and target prices, but also the price of the pattern and compares it with the current price
in the upper picture the first 100 prices (blue) is a trace, then a test...
in the bottom picture the output of the model in the probability of buying
Here's how the picture looks like if the test price is not in the range of the tray
It would be very interesting to develop some signs on this theme, but it's hard to work with absolute prices
Very interesting, honestly.
An interesting observation (strange that no one even thought of it)
If you train the model in a small window of 100 on non-normalized prices, then the model is very good at predicting new data if they (prices) are in the same range as in the tray...
Past prices are important for the model, i.e. levels...
The model fixes not only trailing and target prices, but also the price of the pattern and compares it with the current price
in the upper picture the first 100 prices (blue) is a trace, then a test...
in the bottom picture the output of the model in the probability of buying
This is what the picture looks like if the test price is not in the range of the tray
It would be very interesting to develop some signs on this theme, but it's hard to work with absolute prices
Very interesting, honestly.
We can subtract the moving average and normalize it by volatility. Then you get a stationary series.
You can subtract the moving average and normalize by volatility. Then we get a stationary series.
An interesting observation (strange that no one even thought of it)
If you train the model in a small window of 100 on non-normalized prices, then the model is very good at predicting new data if they (prices) are in the same range as in the tray...
Past prices are important for the model, i.e. levels...
The model fixes not only trailing and target prices, but also the price of the pattern and compares it with the current price
in the upper picture the first 100 prices (blue) is a trace, then a test...
in the bottom picture the output of the model in the probability of buying
Here's how the picture looks like if the test price is not in the range of the tray
It would be very interesting to develop some signs on this theme, but it's hard to work with absolute prices
Very interesting, honestly
I wrote a long time ago from my own observations that normalization is evil. The question is how to minimize it. Because in the trend markets it will always be out of the range and we will have to normalize the features anyway.
Use a general distribution (assuming that price == SB) rather than a sample distribution of a trait for normalization?
Use a general distribution (assuming that price == SB) instead of a sample distribution for normalization?
I'm in favor of some innovative approaches)
I'm all for some innovative approaches )
spin the spinner.
like he told me to do,
you know who I mean?
damn Jew.Spin the spinner.
like he told me to,
you know who you're talking about?
You can't be twitchy, now but later.I wrote a long time ago from my own observations that normalization is evil. The question is how to minimize it. Because in the trend markets there will always be out of the range and we will have to normalize the features anyway.
We can cluster (or in a simpler way divide) prices into ranges, each range can be represented as a stationary series, we can then normalize, and we will remember past prices...
just as an idea...
We can cluster (or in a simpler way divide) prices into ranges, each range can be represented as a stationary series, we can then also normalize, and we will remember past prices...
just as an idea...