import numpy as np
import math


def exponential_moving_average(data, key, alpha=0.5, ic=None):
	result = np.zeros_like(data[key])
	if ic is None:
		r = data[key][0]
	else:
		r = ic
	n = 0
	for v in data[key]:
		r = alpha*r + (1-alpha)*v
		result[n] = r
		n += 1

	return np.array(result)

def moving_average(data, key, window_days, ic=0):
	mem = [ic] * window_days
	result = np.zeros_like(data[key])
	k=0
	cumsum = ic
	n = 0
	for v in data[key]:
		cumsum += (v - mem[k])
		mem[k] = v
		k += 1
		if k >= window_days:
			k=0

		result[n] = cumsum/window_days
		n += 1

	return result

def moving_variance(data, key, window_days, ic=0):
	mem = [ic] * window_days
	result = np.zeros_like(data[key])
	k=0
	cumsum = ic
	data_mean = data[key] - moving_average(data, key, window_days, ic)
	n = 0
	for v in data_mean:
		try:
			v2 = v * v
		except:
			print ('v', v)
		cumsum += (v2 - mem[k])
		mem[k] = v2
		k += 1
		if k >= window_days:
			k=0

		var = max(0, cumsum) / window_days
		result[n] = math.sqrt(var)
		n += 1

	return result

def moving_max(data, key, window_days, ic=-1e9):
	mem = [ic] * window_days
	result = np.zeros_like(data[key])
	k=0
	n = 0
	for v in data[key]:
		mem[k] = v
		k += 1
		if k >= window_days:
			k=0

		result[n] = np.max(mem)
		n += 1

	return result

def moving_min(data, key, window_days, ic=1e9):
	mem = [ic] * window_days
	result = np.zeros_like(data[key])
	k=0
	n = 0
	for v in data[key]:
		mem[k] = v
		k += 1
		if k >= window_days:
			k=0

		result[n] = np.min(mem)
		n += 1

	return result

def normalize(data, key, ic=None):
	result = np.zeros_like(data[key])
	cumsum = 0
	if ic is None:
		prev = data[key][0]
	else:
		prev = ic
	n = 0
	for v in data[key]:
		if prev == 0:
			print (prev)
		cumsum += 100*(v - prev)/(prev)
		result[n] = cumsum

		if v > 0:
			prev = v
		n += 1

	return result


def colsum(data, keys):
	result = np.zeros(data['index'].shape)
	for key in keys:
		result += data[key]

	return result

def macd(data, key):
	ema_short = exponential_moving_average(data, key, alpha=0.85)
	ema_long = exponential_moving_average(data, key, alpha=0.925)

	return ema_short - ema_long

def bollinger(data, window_days, f=2):
	tp = colsum(data, keys=['high', 'low', 'close']) / 3
	tp_dict = {'tp': tp}
	stddev = np.array(moving_variance(tp_dict, 'tp', window_days))

	mid = np.array(moving_average(tp_dict, key='tp', window_days=window_days))
	upper = mid + f * stddev
	lower = mid - f * stddev
	result = {'index' : data['index'], 'upper' : upper, 'mid' : mid, 'lower' : lower}
	return result