I think you have got the main idea more or less correct. The percentages will be subject to tuning.
For me, each candidate is a list of tuples sufficient to cover the whole simulation (until it crashes / lands). The basis of my evaluation is different according to how the simulation of the candidate ends, e.g. fuel remaining if landed, distance to landing area if crashed outside landing area, and angle and velocity if crashed within landing area. There is no single right approach; you can devise your own basis of evaluation.
Hello,
I am facing a timeout issue with Mars Lander Optimization puzzle which I am bit surprised of.
Like in other Optimization puzzles, my bot is considering that on first round 1000 ms is allowed to answer then 100 ms on following turns.
This assumption works well for validators 2 to 5 but not on the first one which is generating a timeout.
If I do submit my code, same thing happens : first test is failing (easiest one) which I guess is for this timeout problem and all the others pass.
Any clue on why this difference between Validator 1 and the others regarding timeout value for first turn ?
Is something specific on Mars Lander Optim puzzle ?
Hi Gloopy. My best guess would be that your simulations take longer in the fist round because there are no mountains close to the starting position. It typically takes longer for the lander to crash in round 1 if you simulate random strategies.
Also, position and speed are stored as floats and rounded at display.
E.g.
if x0 = 1.5 (displayed as x=2)
and vx = 1.9m/s (displayed as 2m/s)
then x1 will be 3.4 (display 3, not 4)
Thanks for your hint WouterSolutions.
I donât think that the sim code is very sensitive to the number of simulations performed, I will check. But you made me realize that I may have a bug elsewhere especially if I cannot land in a fixed number of turns for this case.
After years of hesitation, I decided to give this optimization problem a go. With no prior knowledge of genetic algorithms, the article of Di_Masta was enough to reach the #24 spot. If you want to solve this game but struggle, a simple genetic algorithm does wonders.
Maybe, but iâd have to know what is âSimulated annealingâ and actually do some math, while a monkey-like algo trying out every possible choice and picking the best outcome is easy to do.