Proposal: change/update AI skill values

[shadow-fa]

[See below]

edit: moved refactor discussion to #183. moved
Note that the following discussion might refer to the already refactored ai skill matrix from #183

[osarup]

[edit by shado: moved refactor discussion to #183

Given the changes to the curves that took place in August, it's probably good to revise these values anyway. I'm not sure if a difference between 0.42 and 0.49 is even going to be noticeable. These values haven't been tweaked since 2015 anyway, and a fair few things have changed since.

I'd suggest the following (keeping in mind this is based on intuition + reading around):

f_var_skillSet = [
	//Super	High	Medium	Low
	[ 0.5,	0.4,	0.3,	0.2,	],  // aimingAccuracy
	[ 0.5,	0.4,	0.3,	0.2,	],  // aimingShake
	[ 0.6,	0.4,	0.2,	0,	],  // aimingSpeed
	[ 1,	0.75,	0.5,	0.3,	],  // spotDistance
	[ 0.7,	0.5,	0.4,	0.3,	],  // spotTime
	[ 1,	0.9,	0.7,	0.6,	],  // courage
	[ 1,	1,	0.8,	0.75,	],  // reloadSpeed*
	[ 1,	1,	1,	1,	],  // commanding
	[ 1,	1,	0.8,	0.6,	],  // general
];

*according to 1Tac github + Bear of 1Tac:

reloadSpeed apparently breaks rapid firing in single fire mode for players
iirc if you fired semi very fast it would stop firing

Perhaps why it was set to 1 everywhere, in which case we shouldn't change it.

Anyway, I dare say having 4 levels is useless - just drop Low and rename Super, High, Medium to High, Medium and Low (or something). At present we rarely use Super and probably never use Low.

I had originally thought of making it terrain dependent, for example:

f_var_skillSet = [
	//Open	All	Urban	Jungle
	[ 0.5,	0.4,	0.3,	0.2,	],  // aimingAccuracy
	[ 0.5,	0.4,	0.3,	0.2,	],  // aimingShake
	[ 0.5,	0.4,	0.2,	0,	],  // aimingSpeed
	[ 1,	0.7,	0.4,	0.1,	],  // spotDistance
	[ 0.7,	0.5,	0.4,	0.2,	],  // spotTime
	[ 1,	0.9,	0.6,	0.5,	],  // courage
	[ 1,	1,	0.8,	0.75,	],  // reloadSpeed*
	[ 1,	1,	1,	1,	],  // commanding
	[ 1,	1,	0.6,	0.4,	],  // general
];

I find this more useful (as both a mission maker and a host) to select what difficulty to set it at by default.

[shadow-fa]

4 levels (low,med,high,super):

Anyway, I dare say having 4 levels is useless - just drop Low and rename Super, High, Medium to High, Medium and Low (or something). At present we rarely use Super and probably never use Low.

I disagree. We have quite a few missions that use "Low" by default. If anything "Super" is the one we never use. (I might scan all our missions and check how many missions exactly use which setting.)
However I don't see a problem with keeping all four options, as updating old missions becomes a lot more difficult when removing one option.

terrain dependent options:
Assuming we replace our current options with these, when naively updating a mission it could lead to a completely different and unintended AI behaviour (or even scripting errros due to the different number of options). Additionally it becomes impossible to tell which option makes the AI "better" if an admin should choose to make it harder/easier.
I think it's better to either add them additionally to the other 3-4 options we currently have or just advise (using the wiki) to use "Low" skill for jungle missions.

edit: maybe an additional additive or multiplicative skill vector also makes sense. E.g. one vector for jungle missions that decrease the spotDistance and other skills. This additional vector could then also be tweaked by missionmakers to better tailor the skills to their needs.
But I would argue, simply using the "Low" preset would probably also work sufficiently for these kinds of missions.

changes to skill values:

I'd suggest the following

Just dumping that matrix without any explanation/reasonaning and/or comparison to the old values makes it hard for anyone to understand why we should change anything.
So here's a graphical representation of the suggested changes (differences):

Now that we can see the difference the major (proposed) changes seem to be:

• decrease the aimingSpeed for lower skill settings
• increase the spotDistance for higher skill settings
• (reduce the reloadSpeed for lower skill settings)
• slightly decrease aimingShake (which leads to slightly more weapon sway)
• slightly decrease aimingAccuracy

@NinjaProtocol is this interpretation correct?

If so, I'd like to ask why you want to make these changes.

• why do we want to change courage and general?
• accuracy/shake seems to have been decreased, but is this simply to make the AI worse or because of the engine changes on how these values are interpreted now?
• why do we want such a big change to the aimingSpeed?

Source (sage):

old = matrix([
	[ 0.55,   0.385,   0.33,    0.275,],
	[ 0.6,    0.42,    0.36,    0.3,  ],
	[ 0.6,    0.42,    0.36,    0.3,  ],
	[ 0.7,    0.49,    0.42,    0.35, ],
	[ 0.7,    0.49,    0.42,    0.35, ],
	[ 1,      0.84,    0.72,    0.6,  ],
	[ 1,      1,       1,       1,    ],
	[ 1,      1,       1,       1,    ],
	[ 1,      0.84,    0.72,    0.6,  ],
]);
new = matrix([
	[ 0.5,     0.4,    0.3,      0.2, ],
	[ 0.5,     0.4,    0.3,      0.2, ],
	[ 0.6,     0.4,    0.2,      0,   ],
	[ 1,       0.75,   0.5,      0.3, ],
	[ 0.7,     0.5,    0.4,      0.3, ],
	[ 1,       0.9,    0.7,      0.6, ],
	[ 1,       1,      0.8,      0.75,],
	[ 1,       1,      1,        1,   ],
	[ 1,       1,      0.8,      0.6, ],
]);

def p(m):
    labels=["aimingAccuracy","aimingShake","aimingSpeed","spotDistance","spotTime","courage","reloadSpeed","commanding","general"]
    M = matrix_plot(m,colorbar=True,cmap="coolwarm",colorbar_options={'pad':0.2})
    g = M
    for idx,val in enumerate(labels):
        g += text(val,(M.get_axes_range()['xmax']+0.3,idx),horizontal_alignment='left')
    for idx,val in enumerate(["super","high","medium","low"]):
        g += text(val,(idx,-1.2))
    g.set_axes_range(**M.get_axes_range())
    g.show()

p(new-old)

[SamLex]

Before this goes further, this needs to be split into two issues. There is the original refactor that should not changes any of the final skill values. Then there is the bigger discussion of if the skill values need changed. That is a separate issue.

[SamLex]

Maybe worth looking at Wolfenswan's proposed updated values on the old repo