import random
class equation:
	#equation format:  price delta (%) for a piece of news is equal to S*c^(-kt)
	#S = score, c = constant, k = constant
	c = 1
	k = 1
	def __init__(self,c=1,k=1):
		self.c = c
		self.k = k
	def calc(self,score,dt):
		return score*self.c**(-self.k*dt)
	def calcseries(self, series):
		return sum([self.calc(*dat) for dat in series])
	def __str__(self):
		return "score*%s**(-%s*dt)"%(self.c,self.k)

def regress(target, dataset,generations = 100, children = 8, keep = .5, debug = False):
	"""Regress equations to match target.  Returns the best equation found"""
	import time, re, datetime
	keep = int(str(keep*children).split('.')[0])
	print keep
	m = 0
	expr = re.compile('([0-9]{4})-([0-9]{2})-([0-9]{2}) ([0-9]{2}):([0-9]{2}):([0-9]{2})')
	for item in dataset:
		temp = time.mktime(map(int,expr.match(str(item['time'])).groups())+[0,0,0])/(60*60*24)#timestamp normalized to days
		item['timestamp'] = temp
		if temp > m: m = temp
	for item in dataset:
		item['differential'] = m-item['timestamp']
	eqs = [[0,equation(random.random(),random.random())] for i in range(children)]
	given = []
	for item in dataset:
		given.append([item['normalized_score'],item['differential']])
	for gen in range(generations):
		for eq in eqs:
			try:
				eq[0] = abs(1-target/float(eq[1].calcseries(given)))
			except:
				eq[0] = target
		eqs.sort()
		eqs = eqs[:keep]
		new = []
		for i in range(children-keep):
			parent1 = random.choice(eqs)[1]
			parent2 = random.choice(eqs)[1]
			new.append(spawn(parent1,parent2))
		eqs += new
		if debug:print "Best: %.4f"%eqs[0][0]
	print eqs[0][1].calcseries(given)
	return eqs[0][1]
def spawn(p1, p2, mutchance=.4,mutamt = .2):
	att = random.random()
	r = equation([p1.c,p2.c][att < .5],[p1.k,p2.k][att >= .5])
	if random.random() < mutchance:
		if random.random() < .5:
			r.c = r.c*(1+[mutamt,-mutamt][random.random() < .5])
		else:
			r.k = r.k*(1+[mutamt,-mutamt][random.random() < .5])
	return [0,r]