Large update with mostly rewritten character subsystem

proto2/tests/linreg.gd

@@ -0,0 +1,78 @@
+extends Reference
+class_name LinReg
+
+var data = []
+var m = []
+var b = []
+var errors = []
+var size = 30
+func _init():
+	b.resize(size)
+	m.resize(size)
+	errors.resize(size)
+	for j in range(size):
+		b[j] = 0.0
+		m[j] = 1.0
+	print("hello linreg")
+func sample_x():
+	var ret = []
+	for k in range(size):
+		ret.push_back(randf() * 2.0 - 1.0)
+	return ret
+func iterate(delta):
+	var x = sample_x()
+	var y = 0.0
+	for nx in x:
+		y += nx * nx
+	y /= size
+	data.push_back([x, y])
+	linearRegression_gd()
+#	print(m)
+#	print(b)
+	print(errors)
+	return false
+func linearRegression_sq():
+	var xsum = []
+	var num = []
+	var den = []
+	xsum.resize(size)
+	num.resize(size)
+	den.resize(size)
+	for j in range(xsum.size()):
+		xsum[j] = 0.0
+		num[j] = 0.0
+		den[j] = 0.0
+		b[j] = 0.0
+		m[j] = 1.0
+	var ysum = 0.0
+	for i in range(data.size()):
+		for j in range(data[i][0].size()):
+			xsum[j] += data[i][0][j]
+			ysum += data[i][1]
+	var xmean = []
+	for j in range(xsum.size()):
+		xmean.push_back(xsum[j] / data.size())
+	var ymean = ysum / data.size()
+	for i in range(data.size()):
+		var x = data[i][0]
+		var y = data[i][1]
+		for j in range(x.size()):
+			num[j] += (x[j] - xmean[j]) * (y - ymean)
+			den[j] += (x[j] - xmean[j]) * (x[j] - xmean[j])
+	for j in range(size):
+		m[j] = num[j] / den[j]
+		b[j] = ymean - m[j] * xmean[j]
+
+var learning_rate = 0.15
+func linearRegression_gd():
+	for e in range(30):
+		for i in range(data.size()):
+			var x = data[i][0]
+			var y = data[i][1]
+			for j in range(x.size()):
+				var guess = m[j] * x[j] + b[j]
+				var error = y - guess
+				errors[j] = error
+				m[j] += error * x[j] * learning_rate
+				b[j] += error * learning_rate
+