Merge: Unsolicited proof-of-concept machine learning jackspeed PR #913

etternagame · Oct 30, 2020 · 13f43ed · 13f43ed
2 parents 0c3689c + 5e933e2
commit 13f43ed
Show file tree

Hide file tree

Showing 4 changed files with 32 additions and 16 deletions.
diff --git a/src/Etterna/MinaCalc/Dependent/HD_Sequencers/GenericSequencing.h b/src/Etterna/MinaCalc/Dependent/HD_Sequencers/GenericSequencing.h
@@ -144,9 +144,9 @@ struct Anchor_Sequencing
 			return _len_cap_ms;
 		}
 
-		static const auto avg_ms_mult = 1.075F;
-		static const auto anchor_time_buffer_ms = 25.F;
-		static const auto min_ms = 82.5F;
+		static const auto avg_ms_mult = 1.0363214F;
+		static const auto anchor_time_buffer_ms = 24.336933F;
+		static const auto min_ms = 79.538712F;
 
 		// get total ms
 		const auto total_ms = ms_from(_last, _start);
@@ -167,7 +167,7 @@ struct Anchor_Sequencing
 
 		// BAD TEMP HACK LUL
 		if (_len == 2) {
-			ms *= 1.1F;
+			ms *= 1.1018715F;
 			ms = ms < 155.F ? 155.F : ms;
 		}
 

diff --git a/src/Etterna/MinaCalc/MinaCalc.cpp b/src/Etterna/MinaCalc/MinaCalc.cpp
@@ -105,7 +105,7 @@ Calc::CalcMain(const std::vector<NoteInfo>& NoteInfo,
 		 * files */
 		for (auto i = 0; i < NUM_Skillset; ++i) {
 			if (mcbloop[i] > base * 0.9f) {
-				mcbloop[i] = Chisel(mcbloop[i] * 0.9F,
+				mcbloop[i] = Chisel(mcbloop[i] * 0.82792711F,
 									0.32F,
 									score_goal,
 									static_cast<Skillset>(i),
@@ -298,11 +298,11 @@ JackStamAdjust(const float x, Calc& calc, const int hi)
   -> std::vector<std::pair<float, float>>
 {
 	// Jack stamina Model params (see above)
-	static const auto stam_ceil = 1.05234F;
-	static const auto stam_mag = 23.F;
-	static const auto stam_fscale = 2150.F;
-	static const auto stam_prop = 0.49424F;
-	auto stam_floor = 0.95F;
+	static const auto stam_ceil = 1.0087639F;
+	static const auto stam_mag = 22.086582F;
+	static const auto stam_fscale = 2060.1338F;
+	static const auto stam_prop = 0.48069301F;
+	auto stam_floor = 0.90450478F;
 	auto mod = 0.95F;
 
 	auto avs2 = 0.F;
@@ -316,7 +316,7 @@ JackStamAdjust(const float x, Calc& calc, const int hi)
 	for (size_t i = 0; i < diff.size(); i++) {
 		const auto avs1 = avs2;
 		avs2 = diff.at(i).second;
-		mod += ((((avs1 + avs2) / 2.F) / (stam_prop * x)) - 1.F) / stam_mag;
+		mod += ((((avs1 + avs2) / 2.F) / (stam_prop * x)) - 1.0156831F) / stam_mag;
 		if (mod > 0.95F) {
 			stam_floor += (mod - 0.95F) / stam_fscale;
 		}
@@ -333,12 +333,12 @@ JackStamAdjust(const float x, Calc& calc, const int hi)
 	return doot;
 }
 
-constexpr float magic_num = 16.F;
+constexpr float magic_num = 16.077566F;
 
 [[nodiscard]] inline auto
 hit_the_road(const float& x, const float& y) -> float
 {
-	return std::max(static_cast<float>(magic_num * erf(0.04F * (y - x))), 0.F);
+	return std::max(static_cast<float>(magic_num * erf(0.096623257F * (y - x))), 0.040538613F);
 }
 
 /* ok this is a little jank, we are calculating jack loss looping over the
@@ -516,7 +516,7 @@ Calc::InitializeHands(const std::vector<NoteInfo>& NoteInfo,
  * degree above the actual max points as a cheap hack to water down some of the
  * absurd scaling hs/js/cj had. Note: do not set these values below 1 */
 constexpr float tech_pbm = 1.F;
-constexpr float jack_pbm = 1.0175F;
+constexpr float jack_pbm = 1.0013144F;
 constexpr float stream_pbm = 1.01F;
 constexpr float bad_newbie_skillsets_pbm = 1.05F;
 
@@ -830,6 +830,16 @@ Calc::InitAdjDiff(Calc& calc, const int& hi)
 							 tp_mods[Skill_Jumpstream];
 					*stam_base = max<float>(a, b);
 				} break;
+				case Skill_JackSpeed: {
+					if (i < calc.jack_diff.at(hi).size()) {
+						auto* jack_diff = &calc.jack_diff.at(hi).at(i).second;
+						if (*jack_diff > 0) {
+							auto tech_funk_at_jacks = calc.soap.at(hi).at(TechBase).at(i) * tp_mods.at(ss) * basescalers.at(ss);
+							tech_funk_at_jacks /= max<float>(calc.doot.at(hi).at(CJ).at(i) * calc.doot.at(hi).at(CJ).at(i), 1.0F);
+							*jack_diff = lerp(0.3179549F, *jack_diff, tech_funk_at_jacks / fastsqrt(*jack_diff));
+						}
+					}
+				} break;
 				case Skill_Chordjack:
 					*adj_diff *= fastsqrt(calc.doot.at(hi).at(CJOHJump).at(i));
 					break;
@@ -944,7 +954,7 @@ MinaSDCalcDebug(
 	}
 }
 
-int mina_calc_version = 441;
+int mina_calc_version = 442;
 auto
 GetCalcVersion() -> int
 {

diff --git a/src/Etterna/MinaCalc/PatternModHelpers.h b/src/Etterna/MinaCalc/PatternModHelpers.h
@@ -95,3 +95,9 @@ weighted_average(const float& a, const float& b, const float& x, const float& y)
 {
 	return (x * a + ((y - x) * b)) / y;
 }
+
+inline auto
+lerp(float t, float a, float b) -> float
+{
+	return (1.F - t)*a + t*b;
+}
diff --git a/src/Etterna/MinaCalc/UlbuAcolytes.h b/src/Etterna/MinaCalc/UlbuAcolytes.h
@@ -10,7 +10,7 @@
  * patterns have lower enps than streams, streams default to 1 and chordstreams
  * start lower, stam is a special case and may use normalizers again */
 static const std::array<float, NUM_Skillset> basescalers = {
-	0.F, 0.93F, 0.885F, 0.84F, 0.925F, 0.91F, 0.8F, 0.83F
+	0.F, 0.93F, 0.885F, 0.84F, 0.925F, 0.8833277F, 0.8F, 0.83F
 };
 
 static const std::string calc_params_xml = "Save/calc params.xml";