mirror of
https://github.com/TheAlgorithms/Python.git
synced 2024-11-24 05:21:09 +00:00
7ea812996c
* Adds exponential moving average algorithm * code clean up * spell correction * Modifies I/O types of function * Replaces generator function * Resolved mypy type error * readibility of code and documentation * Update exponential_moving_average.py --------- Co-authored-by: Christian Clauss <cclauss@me.com>
74 lines
2.6 KiB
Python
74 lines
2.6 KiB
Python
"""
|
|
Calculate the exponential moving average (EMA) on the series of stock prices.
|
|
Wikipedia Reference: https://en.wikipedia.org/wiki/Exponential_smoothing
|
|
https://www.investopedia.com/terms/e/ema.asp#toc-what-is-an-exponential
|
|
-moving-average-ema
|
|
|
|
Exponential moving average is used in finance to analyze changes stock prices.
|
|
EMA is used in conjunction with Simple moving average (SMA), EMA reacts to the
|
|
changes in the value quicker than SMA, which is one of the advantages of using EMA.
|
|
"""
|
|
|
|
from collections.abc import Iterator
|
|
|
|
|
|
def exponential_moving_average(
|
|
stock_prices: Iterator[float], window_size: int
|
|
) -> Iterator[float]:
|
|
"""
|
|
Yields exponential moving averages of the given stock prices.
|
|
>>> tuple(exponential_moving_average(iter([2, 5, 3, 8.2, 6, 9, 10]), 3))
|
|
(2, 3.5, 3.25, 5.725, 5.8625, 7.43125, 8.715625)
|
|
|
|
:param stock_prices: A stream of stock prices
|
|
:param window_size: The number of stock prices that will trigger a new calculation
|
|
of the exponential average (window_size > 0)
|
|
:return: Yields a sequence of exponential moving averages
|
|
|
|
Formula:
|
|
|
|
st = alpha * xt + (1 - alpha) * st_prev
|
|
|
|
Where,
|
|
st : Exponential moving average at timestamp t
|
|
xt : stock price in from the stock prices at timestamp t
|
|
st_prev : Exponential moving average at timestamp t-1
|
|
alpha : 2/(1 + window_size) - smoothing factor
|
|
|
|
Exponential moving average (EMA) is a rule of thumb technique for
|
|
smoothing time series data using an exponential window function.
|
|
"""
|
|
|
|
if window_size <= 0:
|
|
raise ValueError("window_size must be > 0")
|
|
|
|
# Calculating smoothing factor
|
|
alpha = 2 / (1 + window_size)
|
|
|
|
# Exponential average at timestamp t
|
|
moving_average = 0.0
|
|
|
|
for i, stock_price in enumerate(stock_prices):
|
|
if i <= window_size:
|
|
# Assigning simple moving average till the window_size for the first time
|
|
# is reached
|
|
moving_average = (moving_average + stock_price) * 0.5 if i else stock_price
|
|
else:
|
|
# Calculating exponential moving average based on current timestamp data
|
|
# point and previous exponential average value
|
|
moving_average = (alpha * stock_price) + ((1 - alpha) * moving_average)
|
|
yield moving_average
|
|
|
|
|
|
if __name__ == "__main__":
|
|
import doctest
|
|
|
|
doctest.testmod()
|
|
|
|
stock_prices = [2.0, 5, 3, 8.2, 6, 9, 10]
|
|
window_size = 3
|
|
result = tuple(exponential_moving_average(iter(stock_prices), window_size))
|
|
print(f"{stock_prices = }")
|
|
print(f"{window_size = }")
|
|
print(f"{result = }")
|