Algorithm Optimisation Championship. - page 122
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
The championship is designed for trained participants, for those who have ready algorithms. I remind you that if you want to read the profile literature beforehand, the link is in my profile and here.
When we have enough trained participants with ready algorithms, then we will have something to talk about. For now, goodbye and see you soon!
I devoted a lot of energy and nerve to the championship, but one moderator can ruin everything with one of his posts. Examples of codes and working algorithm, references to literature, connection interfaces - all this I gave, but it turns out, as the moderator said, all this is not there, I must have imagined it!
If you want to check your algorithms for strength, feel free to contact me in the personal area, and I'll answer your questions.
Don't expect any more public activity from me, please understand me correctly.
With your permission, Andrew, I propose to evaluate it this way, again, I do not insist.
Problem: Find an extremum F(x1,x2, xn) of a function unknown to the algorithm.
Rules:
The winner shall be determined on sum of 2 criteria
Accuracy T=Fmin_best/Fmin_y - (minimum of best/ minimum of participant) in case of finding minimum
T=Fmax_y/Fmax_best in case of searching for the maximum
2. Number of calls to function K=Kbest/Ky (minimum number of calls from all participants)
Total score Os=T+K.
Participant's algorithm passes array double X[x1,x2, xn] and gets the value of the function, then it operates according to its logic and calculates the extremum and the number of calls (number of passes of the array).
The search range is declared separately.
We can take the function F(x1,x2,x3)=exp(x1+x2+x3)/(x1*x2*x2*x3*x3) for training;
My algorithm got these results
I got these results.
The specified search error - 0.01 Initial parameters (first call) x1=x2=x3=0.5; The search range 0-100
Number of times the function is called Ky=51
Minimum Fmin_y=3.76210
x1=1.1; x2=2.1; x3=3.1;
With your permission, Andrew, I propose to evaluate it this way, again, I do not insist.
Problem: Find an extremum F(x1,x2, xn) of a function unknown to the algorithm.
Rules:
The winner shall be determined on sum of 2 criteria
Accuracy T=Fmin_best/Fmin_y - (minimum of best/ minimum of participant) in case of finding minimum
T=Fmax_y/Fmax_best in case of searching for the maximum
2. Number of calls to function K=Kbest/Ky (minimum number of calls from all participants)
Total score Os=T+K.
Participant's algorithm passes array double X[x1,x2, xn] and gets the value of the function, then it operates according to its logic and calculates the extremum and the number of calls (number of passes of the array).
The search range is declared separately.
We can take the function F(x1,x2,x3)=exp(x1+x2+x3)/(x1*x2*x2*x3*x3) for training;
My algorithm got these results
I got these results.
The specified search error - 0.01 Initial parameters (first call) x1=x2=x3=0.5; The search range 0-100
Number of times the function is called - Ky=51
Minimum Fmin_y=3.76210
x1=1.1; x2=2.1; x3=3.1;
Here I have shown you how places are calculated. The only clarification is not the time, but the number of calls to the FF:
Andrey Dik:Here is an example of the calculation of the championship place for the competitors. The table is already sorted by the "General Criterion" column.
Let's take a closer look at the table. Petrov scored 0.89 for the "Accuracy" criterion, where the maximum FF value scales to 1.0 and the minimum to 0.0. It is thus the best value of all. Zaborov had the lowest score on this criterion, he managed to score only 0.21 and it already seemed that Zaborov would be sitting on the fence in the standings....
But according to the "Time" criterion, Zaborov is the leader - he is the fastest algorithm and completed the task in 221 seconds (but not the most inexact one at the same time).
Now let's calculate general criterion. And what do we see? Who would have thought it, but Abamov is in last place, because his overall criterion is the lowest, he could only score 0.53.
ZS. In the accuracy criterion, the highest value is better; in the time criterion, the lowest value is better. All the coincidences of the names are accidental, for me it was a great surprise to see Abamov get the last place, well, you'll have to be more precise... and faster.
ZZZI. To calculate the "Accuracy" criterion the formula(In - InMIN) / (InMAX - InMIN) is applied; and to calculate the "Time" criterion(InMIN - In) / (InMAX - InMIN);
PS. How many tests do you do in your task? Why initialize with a fixed value of 0.5? How many accesses do you get if you initialise with a number of 10.0?
PPS..... any questions, please send them to the personal section.
This is where I showed how participant places are calculated. The only clarification is not the time, but the number of references to the FF:
PS. How many trials are you doing in your task? Why initialise with a fixed value of 0.5? How many accesses do you get if you initialize with number 10.0?
G When initializing 10.0 you get a number of hits of 2028 at least 3.73722. I have not checked yet at initialization of RSCh.
Understand I am not claiming at all that my algorithm is better than yours, I am still working on mine. I just wanted to know how the total score will be formed. How to sharpen the algorithm.
You are right, I am the only one who says I have something ready. Other participants are publicly silent. It is reasonable to postpone the championship until the autumn and train in the meantime.
G When initialising 10.0 the number of hits is 2028 at least 3.73722. I have not yet checked the initialization of the RNG.
Understand I do not claim that my algorithm is better than yours, I am still working on mine. I just wanted to know how the total score will be formed. How to sharpen the algorithm.
ZS. You're right, it seems that only I say that I have something ready. Other participants are publicly silent. It is reasonable to postpone the championship to the autumn, and in the meantime to train.
Well, that's what I thought.
That's the thing. You don't have to tweak it to fit anything. The algorithm should be able to work with any functions not known to it, and any initialization is equivalent, and you need to do several tests, at least 20.
Put the communication in the private.
My grandfather used to tell me: "The difference between a wise man and a fool is that he gets things done.
Didn't your grandfather tell you... "don't interfere with the busybody"?
Are you busy?
Although, you are a classic of genre - how to talk about championship for two months, get participants together, get support from MC in the form of $3000 prize money, but then you blurt it out and get nothing.
Pushkin's old lady with a leaky trough was luckier in the end -- at least she got her trough back.
Are you busy?
Although, you are a classic of genre - how to talk about championship for two months, get participants together, get support from MC in the form of $3000 prize money, but then you blurt it out and get nothing.
Pushkin's old lady with a leaky trough was luckier in the end - at least she got her trough back.