Market etiquette or good manners in a minefield - page 88
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
Tell me, how do you organise reading of records in MQL?
I.e., there is a file with tick history, which is constantly updated. The RT should be built based on data stored in this file - i.e., ticks, but the Expert Advisor (the grid) does not need ticks itself, it only needs RT oscillations. It means that check of tick history for readiness of a new reference should be done every tick... it's a bit expensive, but I don't see any other way yet. Maybe it would be more reasonable to organize a parallel work of two EAs, one of which only breaks tick history and informs the other one about readiness of a new tick... Then the second EA should check value of global variable at every tick and start working only when new datum arrives.
I'm just about to switch to ticks. Working with a number of Open M1's for now.
It's happening! I have four different grids giving the results of statistical modelling of EURUSD 1h trades, number of training epochs 1000:
The figure on the left shows profitability as the average of pips per transaction (averaged over 20 independent numerical experiments). The red one shows the performance of a single linear neuron as a function of the number of inputs (abscissa axis). Blue - non-linear NS with one hidden layer and two neurons in it (output of 1 neuron - Buy/Sell). Black - with four neurons in hidden layer and lilac - with 8. You can see that with increasing number of inputs, the yield slowly increases for all configurations, and with increasing number of neurons in hidden layer the stability of NS-based NS slightly increases. To the right are the normalized variance plots for the training sample (with index P) and for the test sample (with index E). The normalisation was performed on the variance of the input data. A value of <1 indicates that the network is "trained". For all configurations, the training sample length P was assumed to be P=w^2/d. The fact that for NS with small number of neurons there is a strong divergence of variance of training and test samples, indicates a small sample length according to this estimation, and on the contrary, at the number of neurons more than 4 in the hidden layer, we observe the convergence of these indicators, which indicates an overestimation of the length. In general, we can state the invalidity of the given estimation and the problem of its correct solution remains open and highly relevant! Mathemat, how are you doing on this issue?
Judging by your graphs, the results of the single layer are quite good!
And what, if not a secret, is the abscissa of the maximum for the lilac line (picture on the left)... does it look like an 8?
Fedor, they are, above all, unstable for a single layer and lead to inflated risks that we have to take on ourselves in the end!
I have now loaded my statistical MACHINE to analyse 2000 epochs.
Let's compare... What will we do if it turns out that there is money to be made on watches?
I suggest renting a high-performance server for statistical calculations, because it is unrealistic for me to keep my computer on 24 hours a day. And there are higher speeds and calculations can be made round the clock. It costs just 950 rubles. a month. Access as a remote desktop - very convenient. What do you think?
And what, if it's no secret, is the abscissa of the maximum for the lilac line (picture on the left)... looks like an 8?Like yes - 2^3=8 inputs. This data should be treated as preliminary.
And what is the performance of this server? I suppose it's 10-50 times less than a good home machine, taking into account that an average of 1000-10000 users are connected to it round-the-clock! So what's the gain? My computer hasn't shut down for weeks... I'm used to it. I'm used to it now.
I don't know the performance, and the clock speed is 1Ghz, but the server is dedicated, there will be no one on it but you.
For me, the advantage is that I have a 24/7 computer, 24/7 high-speed internet, and all this in a specially equipped room and not at my home.
I do for the collection of ticks and a virtual servavkom, and it's cheaper by half.
It's here: Internet industry
...but it's a dedicated server, there's no one else on it but you.
I think that's what they tell everyone who sits on that server:-)