j132 strafe pick quality: chosen tile's heat in the [strafe] log line (heat=h/10.0 alt=best-other ok=0|1) and ON the tile (green/yellow/red ramp + value), legible with TR_STRAFE_HEAT_GRID=0; deterministic display check vs pathMaxHeat
This commit is contained in:
@@ -0,0 +1,157 @@
|
||||
## Deterministic check for the j132 STRAFE pick-quality DISPLAY.
|
||||
##
|
||||
## NO battle, NO Java, NO server. Run with:
|
||||
## nim c -r --nimcache:/tmp/nc_j132 --path:common_libs \
|
||||
## common_libs/tests/measure_strafe_heat_display.nim
|
||||
##
|
||||
## It drives `computeMove` over the recorded DrussGT fixture (the SAME stream
|
||||
## `measure_strafe_gates.nim` replays) and asserts, for every tick with a valid
|
||||
## target, that the number the display shows (`targetPathHeat`) is the field's
|
||||
## REAL reading for that tile:
|
||||
##
|
||||
## pathMaxHeat(bot -> chosen tile centre)
|
||||
##
|
||||
## On every PICK it also asserts that the number the `[strafe]` line prints
|
||||
## (`pickPathHeat`, formatted by `strafeHeatText`) is that same number, and that
|
||||
## a pick at or under `PathDangerThreshold` is labelled safe (`ok=1`), i.e. that
|
||||
## a "pick" (safe pool non-empty) never carries a heat above the threshold.
|
||||
##
|
||||
## This is a DISPLAY check. It changes nothing the mover decides: it only proves
|
||||
## the displayed/logged heat is the tile's actual path heat.
|
||||
|
||||
import std/[os, json, math, random, strformat]
|
||||
import std/strutils except fromHex # `fromHex` clashes with color.fromHex
|
||||
import gun_harness/gun_interface
|
||||
# Private-field access: include (do NOT import) the mover, exactly as
|
||||
# `test_tfil_commit_env.nim` includes `the_floor_is_lava.nim`.
|
||||
include movements/strafe
|
||||
|
||||
const repoRoot = currentSourcePath().parentDir.parentDir.parentDir
|
||||
const fixtureRel = "tr_drussgt_vs_modularbot.jsonl"
|
||||
const Seed = 20250923
|
||||
const ArenaW = 800.0
|
||||
const ArenaH = 600.0
|
||||
|
||||
var failures = 0
|
||||
proc check(name: string, ok: bool) =
|
||||
if ok: echo "PASS: ", name
|
||||
else: echo "FAIL: ", name; inc failures
|
||||
|
||||
proc loadStates(): seq[WorldState] =
|
||||
let path = repoRoot / "tools" / "fixtures" / fixtureRel
|
||||
for rawLine in lines(path):
|
||||
let line = rawLine.strip()
|
||||
if line.len == 0: continue
|
||||
let n = parseJson(line)
|
||||
if n.hasKey("meta") or n.hasKey("end"): continue
|
||||
let ex = n["ex"].getFloat()
|
||||
let ey = n["ey"].getFloat()
|
||||
result.add WorldState(
|
||||
enemyX: ex, enemyY: ey,
|
||||
enemyHeading: n["eh"].getFloat(), enemySpeed: n["es"].getFloat(),
|
||||
enemyEnergy: n["ee"].getFloat(),
|
||||
selfX: n["sx"].getFloat(), selfY: n["sy"].getFloat(),
|
||||
selfHeading: n["sh"].getFloat(), selfSpeed: n["ss"].getFloat(),
|
||||
selfEnergy: n["se"].getFloat(),
|
||||
arenaWidth: ArenaW, arenaHeight: ArenaH,
|
||||
tick: n["tick"].getInt(),
|
||||
enemies: @[EnemyInfo(id: 1, x: ex, y: ey,
|
||||
heading: n["eh"].getFloat(), speed: n["es"].getFloat(),
|
||||
energy: n["ee"].getFloat())])
|
||||
|
||||
# ── the replay ───────────────────────────────────────────────────────────────
|
||||
|
||||
proc main() =
|
||||
let states = loadStates()
|
||||
check fmt"fixture loaded ({states.len} ticks)", states.len > 0
|
||||
randomize(Seed)
|
||||
var m = initStrafe()
|
||||
m.debugGraphics = false
|
||||
|
||||
var
|
||||
ticks = 0
|
||||
targetTicks = 0
|
||||
picks = 0
|
||||
under = 0 ## picks labelled ok=1 (pathHeat <= PathDangerThreshold)
|
||||
over = 0 ## picks labelled ok=0 (forced fallback/escape)
|
||||
comfy = 0 ## picks <= HALF the threshold (comfortably safe)
|
||||
altKnown = 0 ## picks where a "best alternative" existed
|
||||
dispBad = 0 ## display value != real pathMaxHeat of the chosen tile
|
||||
pickBad = 0 ## picker value != real pathMaxHeat of the chosen tile
|
||||
labelBad = 0 ## a "pick" (safe pool non-empty) with heat > threshold
|
||||
textBad = 0 ## the printed text does not carry the same number
|
||||
minH = Inf
|
||||
maxH = -Inf
|
||||
sumH = 0.0
|
||||
bandComfy = 0 ## <= 2.5 (half of half)
|
||||
bandLow = 0 ## 2.5 .. 5.0
|
||||
bandMarginal = 0 ## 5.0 .. 10.0
|
||||
|
||||
for ws in states:
|
||||
discard m.computeMove(ws)
|
||||
inc ticks
|
||||
if not m.targetValid: continue
|
||||
inc targetTicks
|
||||
|
||||
# The field's REAL reading for the chosen tile, this tick.
|
||||
let realHeat = m.pathMaxHeat(ws.selfX, ws.selfY, m.targetX, m.targetY)
|
||||
if abs(realHeat - m.targetPathHeat) > 1e-9: inc dispBad
|
||||
|
||||
# A pick happened on this tick when the picker reset `lastPickCall` to the
|
||||
# pre-increment `callCount` (computeMove increments callCount last).
|
||||
if m.callCount != m.lastPickCall + 1: continue
|
||||
inc picks
|
||||
if abs(m.pickPathHeat - realHeat) > 1e-9: inc pickBad
|
||||
|
||||
let safe = strafeHeatSafe(m.pickPathHeat)
|
||||
if safe: inc under else: inc over
|
||||
if m.pickPathHeat <= PathDangerThreshold * 0.5: inc comfy
|
||||
if m.pickPathHeat <= PathDangerThreshold * 0.25: inc bandComfy
|
||||
elif m.pickPathHeat <= PathDangerThreshold * 0.5: inc bandLow
|
||||
elif safe: inc bandMarginal
|
||||
if m.targetBestAlt >= 0.0: inc altKnown
|
||||
if m.pickPathHeat < minH: minH = m.pickPathHeat
|
||||
if m.pickPathHeat > maxH: maxH = m.pickPathHeat
|
||||
sumH += m.pickPathHeat
|
||||
|
||||
# A pick from the safe pool can never be over the threshold.
|
||||
if m.lastMode == "pick" and not safe: inc labelBad
|
||||
|
||||
# The EXACT string the log line prints for this pick, and the exact string
|
||||
# the GUI prints on the chosen tile, must carry that same number.
|
||||
let pickTxt = strafeHeatText(m.pickPathHeat, m.targetBestAlt)
|
||||
let okTxt = if safe: "ok=1" else: "ok=0"
|
||||
if not pickTxt.startsWith(fmt"heat={realHeat:.1f}/{PathDangerThreshold:.1f}"):
|
||||
inc textBad
|
||||
if not pickTxt.contains(okTxt): inc textBad
|
||||
if not strafeHeatText(m.targetPathHeat, m.targetBestAlt).startsWith(
|
||||
fmt"heat={m.targetPathHeat:.1f}/"):
|
||||
inc textBad
|
||||
|
||||
check fmt"every displayed heat == pathMaxHeat of the chosen tile " &
|
||||
fmt"({targetTicks} targeted ticks, {dispBad} mismatches)", dispBad == 0
|
||||
check fmt"every printed pick heat == the picker's real path heat " &
|
||||
fmt"({picks} picks, {pickBad} mismatches)", pickBad == 0
|
||||
check fmt"a safe pick is labelled safe and never exceeds the threshold " &
|
||||
fmt"({under} under / {over} over, {labelBad} bad)", labelBad == 0
|
||||
check fmt"`strafeHeatText` carries the displayed number ({textBad} bad)",
|
||||
textBad == 0
|
||||
check "the stream contains picks to check", picks > 0
|
||||
|
||||
echo ""
|
||||
echo fmt"replayed {ticks} ticks, {targetTicks} with a target, {picks} picks"
|
||||
echo fmt"labelled: under={under} over={over} comfortably-safe(<= " &
|
||||
fmt"{PathDangerThreshold * 0.5:.1f})={comfy}"
|
||||
echo fmt"chosen pathHeat: min={minH:.1f} mean={sumH / max(1.0, picks.float):.1f} " &
|
||||
fmt"max={maxH:.1f} best-alt known={altKnown}/{picks}"
|
||||
echo fmt"buckets: <=2.5 {bandComfy} | 2.5-5 {bandLow} | 5-10 {bandMarginal} | " &
|
||||
fmt">10 (forced) {over}"
|
||||
echo fmt"sample log fields: [{strafeHeatText(7.3, 9.8)}] " &
|
||||
fmt"[{strafeHeatText(11.4, -1.0)}]"
|
||||
echo ""
|
||||
if failures > 0:
|
||||
echo failures, " check(s) FAILED"
|
||||
quit(1)
|
||||
echo "All STRAFE pick-quality display checks passed."
|
||||
|
||||
main()
|
||||
Reference in New Issue
Block a user