Merc fastest in S1, third in S2 and second in S3
- Login or Register
No account yet? Sign up
That is what old school guys did in the past but nowadays they are using A.I. methods like Ferrari is doing now with their Reinforcement Learning A.I. software, which could handle multiple variables and parameters at once in real-time...Rodak wrote: ↑23 Mar 2024, 01:26It's 'hone in'! But seriously, changing more than one variable at a time means you have no way to understand how each variable affects the result and you are indeed taking stabs in the dark.I suggest they try to hoan in on the best performing set-up every weekend, step by step, not getting ahead of themselves. Then try to find the common denominator between the best performing set-ups and apply that knowledge to the simulator. The issue I have is these "experiments" seem more like stabs in the dark. It's trying to take three steps at once and more often than not (if not always) they end up back on square one.
Things that can be tried during free practice are limited but add up to many possible changes. The team can quickly change ride height, camber and pressure (to Pirelli limits), toe, wings, shock (damper) settings, sway bars, suspension interactions, engine modes, etc, so there are an enormous number of combinations. This is where the step by step approach is required. Radical changes will tell you nothing. If we had unlimited testing these issue would probably be quickly solved; simulators are great, but they are not real life, especially with aero components when there are correlation issues.
AI and machine learning aren't magic. You still need to train them and give them good data for them to give good results.atanatizante wrote:That is what old school guys did in the past but nowadays they are using A.I. methods like Ferrari is doing now with their Reinforcement Learning A.I. software, which could handle multiple variables and parameters at once in real-time...Rodak wrote: ↑23 Mar 2024, 01:26It's 'hone in'! But seriously, changing more than one variable at a time means you have no way to understand how each variable affects the result and you are indeed taking stabs in the dark.I suggest they try to hoan in on the best performing set-up every weekend, step by step, not getting ahead of themselves. Then try to find the common denominator between the best performing set-ups and apply that knowledge to the simulator. The issue I have is these "experiments" seem more like stabs in the dark. It's trying to take three steps at once and more often than not (if not always) they end up back on square one.
Things that can be tried during free practice are limited but add up to many possible changes. The team can quickly change ride height, camber and pressure (to Pirelli limits), toe, wings, shock (damper) settings, sway bars, suspension interactions, engine modes, etc, so there are an enormous number of combinations. This is where the step by step approach is required. Radical changes will tell you nothing. If we had unlimited testing these issue would probably be quickly solved; simulators are great, but they are not real life, especially with aero components when there are correlation issues.
I suggest you look at the three body problem.....atanatizante wrote: ↑23 Mar 2024, 05:40That is what old school guys did in the past but nowadays they are using A.I. methods like Ferrari is doing now with their Reinforcement Learning A.I. software, which could handle multiple variables and parameters at once in real-time...Rodak wrote: ↑23 Mar 2024, 01:26It's 'hone in'! But seriously, changing more than one variable at a time means you have no way to understand how each variable affects the result and you are indeed taking stabs in the dark.I suggest they try to hoan in on the best performing set-up every weekend, step by step, not getting ahead of themselves. Then try to find the common denominator between the best performing set-ups and apply that knowledge to the simulator. The issue I have is these "experiments" seem more like stabs in the dark. It's trying to take three steps at once and more often than not (if not always) they end up back on square one.
Things that can be tried during free practice are limited but add up to many possible changes. The team can quickly change ride height, camber and pressure (to Pirelli limits), toe, wings, shock (damper) settings, sway bars, suspension interactions, engine modes, etc, so there are an enormous number of combinations. This is where the step by step approach is required. Radical changes will tell you nothing. If we had unlimited testing these issue would probably be quickly solved; simulators are great, but they are not real life, especially with aero components when there are correlation issues.
Narrow performance window. Was an issue with the W13 and W14 tooHammerfist wrote: ↑23 Mar 2024, 08:12That car is bonkers man. Lewis was faster in fp3 than he was in qualifying. George only about a tenth faster in quali. Wtf is going on?
organic wrote: ↑23 Mar 2024, 08:18Narrow performance window. Was an issue with the W13 and W14 tooHammerfist wrote: ↑23 Mar 2024, 08:12That car is bonkers man. Lewis was faster in fp3 than he was in qualifying. George only about a tenth faster in quali. Wtf is going on?
