Market etiquette or good manners in a minefield - page 55
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
to Neutron.
Вот фраер!
Do you think I know less strong expressions? :о)
Where did you see me talking about ACF? No need to make things up here and answer yourself with your own drawings.
Yeah, and your stupid religion doesn't allow you to see the first value in the chart? I told you:
What's the high correlation here?
Yeah, sometimes it can be as high as 0.3-0.4 on the first leg.
Can't you read? And you think it's high?
You call that a weak correlation? To predict such a series you just need to know the sign and amplitude of the last count and that's it!
Then take a prediction AR, just be honest. You don't know that it will NOT predict, that's not enough, you'll get about 50/50. And I sequentially transform a number of times before I make a prediction, then rebuild it. It's not that simple. Sprgnosticate and see.
I won't even say anything about inevitable FZ - I'm already sick of it in your execution. You can ask Privalych for order, he will explain everything in simple language, like for a tank driver :-). He has enough patience!
You and Prival are alike ..... You got phase lag here. Fucking clever fuckers! Here's a picture of one bar, stolen from a colleague from a neighbouring thread:
What's your (H+L)/2 phase lag? From Close? From High? From Low? One bar - one timeframe. This time frame corresponds to
These figures and series are EQUAL, the price has visited them all, and TIME HAS BEEN ONE!!! But predicting (H+L)/2 has a better chance of success, not because correlation (not high), but because it is the average in the bar
You don't seem to understand at all what others are writing about, you're just listening to yourself puffed up. I'm not a moving average that does have a phase delay, I've chosen the series I'm working with. Understand you DHHHHHHB, the row chosen is (H+L)/2, not MA. This row (H+L)/2 is NOT ANY different from Closr, Open etc. IT IS NOT LATE FOR ANYTHING. NOTHING!!!!!! It's your phase delay ... in your head.
Addendum: Just to be clear. These figures are equivalent because they are obtained by processing a series "inside" that bar, i.e. a series bounded by the same time frames. Other parts of the price series (on the sides) are not taken into account. I do not know how to explain it better
Well, I've already written it.
You'd better explain to this nugget why it happens, because I am helpless.
I am not a mathematician, and I am calm about it, I am learning and am not ashamed of it, I am no match for some ... who know everything :o)
I can predict like that too, by the way. Even without an AR. :) But it didn't give me anything. I can guess "where" the price will go with an accuracy of 80% but I have no PROFITS. It's sad. ;)
OK, I'll formulate my idea in more detail tonight.
These figures and series are EQUAL, the price visited them all, and TIME HAS BEEN ONE FOR ALL!!! But only by predicting (H+L)/2 there is more chance of successful forecasts, not because of correlation (not high), but because it is the average value in a bar
These numbers are equal relative to each other, they are. However, their lag is different. They also have different times.
Close -- 0
High -- 0.5
Low -- 0.5
Open -- 1
(High + Low)/2 -- 0.5
(Open + Close)/2 -- 0.5
(H + L + C + C)/4 -- 0.25
This row (H+L)/2 is NO DIFFERENT from Closr, Open etc. IT IS NOT LATE FOR ANYTHING. NOTHING!!!!!! It's your phase delay ... in your head.
Look, Sergei, I'll explain it to you one more time on my fingers!
For certainty we build a synthetic BP of Bernoulli type (when increments are equal in amplitude and random in sign) from 1000 samples (green line). Let's form bars in 100-samples by finding maxima and minima in the interval of 100 left from the current moment, and so on each sample:
Find the highs and lows and plot their half sum (black line).
Now, if you're not blind, what you see with your eyes on the above chart is called smoothing, and the lag of this muving from the quotient is a FZ.
Lest you yell here about my programming skills and head, I'm laying out the code. So go and bang your head on something hard yourself! Because you understand that it's not about scramble, it's about DNA:-)
It's not that simple. I downloaded it, but I can't put it in. I ordered a disc with all the Matcads from one office by post.
OK, for now I'll get on with polishing the pass on the single ply - it can't hurt. By the way, I wanted to ask:
If I feed a binary vector to a single layer and only want a sign on the output... how do I achieve that? It tries to guess the amplitude anyway
If I find my installation disk, I'll post it.
If I feed a binary vector to a single layer and only want a sign on the output... how do I achieve this? It's still trying to guess the amplitude.
Already when NS has given a predicted value (a real number), take its sign using sign(x) function - you will get the desired +/-1.
Ok! Tell me if it's not possible to teach the grid a fixed number of epochs, but, say, until a certain minimum error is reached. Finding the optimal number of epochs, inputs and temperatures is quite a cumbersome and time-consuming task.
Oh, and I almost forgot:
This already applies to valid inputs. How do you calculate the MO of the input vector?