#!/usr/bin/env python
# On 20130704 by lopezdeprado@lbl.gov
#----------------------------------------------------------------------------------------
def testAccuracy_MC(sr_base,sr_case):
	# Test the accuracy of CSCV against hold-out
	# It generates numTrials random samples and directly computes the …
	# … proportion where OOS performance was below the median.
	length,numTrials,numMC=1000,100,1000
	pathOutput='H:/Studies/Quant #23/paper/'
	#1) Determine mu,sigma
	mu_base,sigma_base=sr_base/(365.25*5/7.),1/(365.25*5/7.)**.5
	mu_case,sigma_case=sr_case/(365.25*5/7.),1/(365.25*5/7.)**.5
	hist,probOverfit=[],0
	#2) Generate trials
	for m in range(numMC):
		for i in range(1,numTrials):
			j=np.array([gauss(0,1) for j in range(length)])
			j*=sigma_base/np.std(j) # re-scale
			j+=mu_base-np.mean(j) # re-center
			j=np.reshape(j,(j.shape[0],1))
			if i==1:pnl=np.copy(j)
			else:pnl=np.append(pnl,j,axis=1)
		#3) Add test case
		j=np.array([gauss(0,1) for j in range(length)])
		j*=sigma_case/np.std(j) # re-scale
		j+=mu_case-np.mean(j) # re-center
		j=np.reshape(j,(j.shape[0],1))
		pnl=np.append(pnl,j,axis=1)
		#4) Run test
		# Reference distribution
		mu_is=[np.average(pnl[:length/2,i]) for i in range(pnl.shape[1])]
		sigma_is=[np.std(pnl[:length/2,i]) for i in range(pnl.shape[1])]
		mu_oos=[np.average(pnl[length/2:,i]) for i in range(pnl.shape[1])]
		sigma_oos=[np.std(pnl[length/2:,i]) for i in range(pnl.shape[1])]
		sr_is=[mu_is[i]/sigma_is[i] for i in range(len(mu_is))]
		sr_oos=[mu_oos[i]/sigma_oos[i] for i in range(len(mu_oos))]
		print m,sr_is.index(max(sr_is)),max(sr_is), \
			sr_oos[sr_is.index(max(sr_is))]
		sr_oos_=sr_oos[sr_is.index(max(sr_is))]
		hist.append(sr_oos_)
		if sr_oos_<np.median(sr_oos):probOverfit+=1
	probOverfit/=float(numMC)
	print probOverfit
	return