refactor docs

optimization/WSSAlgorithms.py

@@ -17,6 +17,7 @@ class WindowSizeSelection:
  type of WSS algorithm. It can be 'highest_autocorrelation', 'dominant_fourier_frequency',
  'summary_statistics_subsequence' or 'multi_window_finder'.
  By default, it is 'dominant_fourier_frequency'
+
  Reference:
  (c) "Windows Size Selection in Unsupervised Time Series Analytics: A Review and Benchmark. Arik Ermshaus,
  Patrick Schafer, and Ulf Leser"
@@ -53,10 +54,10 @@ def __init__(self,
  self.window_max = len(time_series)
 
  def autocorrelation(self):
- """
- Main function for the highest_autocorrelation method
+ """ Main function for the highest_autocorrelation method
 
- :return: a tuple of selected window size and list of scores for this method
+ Return:
+ a tuple of selected window size and list of scores for this method
  """
  list_score = [self.high_ac_metric(self.time_series[i:] + self.time_series[:i], i) \
  for i in range(self.window_min, self.window_max)]
@@ -65,11 +66,15 @@ def autocorrelation(self):
 
  def high_ac_metric(self, copy_ts, i):
  """
- Calculate metric value based on chosen chosen window size for the highest_autocorrelation method
+ Calculate metric value based on chosen window size for the highest_autocorrelation method
+
 
- :param copy_ts: a list of lagged time series
- :param i: temporary window size (or the lagged value)
- :return: score for selected window size
+ Args:
+ copy_ts: a list of lagged time series
+ i: temporary window size (or the lagged value)
+
+ Return:
+ score for selected window size
  """
  temp_len = len(copy_ts)
  temp_coef = 1 / (temp_len - i)
@@ -80,11 +85,13 @@ def high_ac_metric(self, copy_ts, i):
  return a_score
 
  def local_max_search(self, score_list):
- """
- Find global max value id for the highest_autocorrelation method.
+ """ Find global max value id for the highest_autocorrelation method.
 
- :param score_list: a list of scores obtained
- :return: a tuple of window_size_selected and list_score
+ Arg:
+ score_list: a list of scores obtained
+
+ Return:
+ a tuple of window_size_selected and list_score
  """
  list_probably_peaks = find_peaks(score_list)[0][1:]
  list_score_peaks = [score_list[i] for i in list_probably_peaks]
@@ -93,10 +100,10 @@ def local_max_search(self, score_list):
  return window_size_selected, list_score_peaks
 
  def dominant_fourier_frequency(self):
- """
- Main function for the dominant_fourier_frequency
+ """ Main function for the dominant_fourier_frequency
 
- :return: a tuple of window_size_selected and list_score
+ Return:
+ a tuple of window_size_selected and list_score
  """
  list_score_k = []
  for i in range(self.window_min, self.window_max):
@@ -107,11 +114,13 @@ def dominant_fourier_frequency(self):
  return window_size_selected, list_score_k
 
  def coeff_metrics(self, temp_size):
- """
- Find score coefficient for the dominant_fourier_frequency
+ """ Find score coefficient for the dominant_fourier_frequency
 
- :param temp_size: temporary selected window size
- :return: a score metric distance
+ Arg:
+ temp_size: temporary selected window size
+
+ Return:
+ score metric distance
  """
 
  length_n = len(self.time_series)
@@ -125,7 +134,8 @@ def summary_statistics_subsequence(self):
  """
  Main function for the summary_statistics_subsequence
 
- :return: selected window size and a list of score
+ Return:
+ selected window size and a list of score
  """
  ts = (self.time_series - np.min(self.time_series)) / (np.max(self.time_series) - np.min(self.time_series))
 
@@ -136,14 +146,16 @@ def summary_statistics_subsequence(self):
  return window_size_selected, list_score
 
  def stats_diff(self, ts, window_size, stats_ts):
- """
- Find difference between global statistic and statistic of subsequnces with different window
+ """ Find difference between global statistic and statistic of subsequnces with different window
  for the summary_statistics_subsequence
 
- :param ts: time series data
- :param window_size: temporary selected window size
- :param stats_ts: statistic over all ts for calculations
- :return: not normalized euclidian distance between selected window size and general statistic for ts
+ Args:
+ ts: time series data
+ window_size: temporary selected window size
+ stats_ts: statistic over all ts for calculations
+
+ Return:
+ not normalized euclidian distance between selected window size and general statistic for ts
  """
  stat_w = [[np.mean(ts[i:i + window_size]), np.std(ts[i:i + window_size]),
  np.max(ts[i:i + window_size]) - np.min(ts[i:i + window_size])] for i in range(self.length_ts)]
@@ -153,24 +165,26 @@ def stats_diff(self, ts, window_size, stats_ts):
  return np.mean(stat_diff)
 
  def suss_score(self, ts, window_size, stats_ts):
- """
- Find score coefficient for the the summary_statistics_subsequence
+ """ Find score coefficient for the summary_statistics_subsequence
+
+ Args:
+ ts: time series data
+ window_size: temporary selected window size
+ stats_ts: statistic over all ts for calculations
 
- :param ts: time series data
- :param window_size: temporary selected window size
- :param stats_ts: statistic over all ts for calculations
- :return: normalized euclidian distance between selected window size and general statistic for ts
+ Return:
+ normalized euclidian distance between selected window size and general statistic for ts
  """
  s_min, s_max = self.stats_diff(ts, len(ts), stats_ts), self.stats_diff(ts, 1, stats_ts)
  score = self.stats_diff(ts, window_size, stats_ts)
  score_normalize = (score - s_min) / (s_max - s_min)
  return 1 - score_normalize
 
  def multi_window_finder(self):
- """
- Main function for multi_window_finder method
+ """ Main function for multi_window_finder method
 
- :return: selected window size and a list of scores for this method
+ Return:
+ selected window size and a list of scores for this method
  """
  distance_scores = [self.mwf_metric(i) for i in range(self.window_min, self.window_max)]
  minimum_id_list, id_max = self.top_local_minimum(distance_scores)
@@ -179,11 +193,13 @@ def multi_window_finder(self):
  return window_size_selected, distance_scores
 
  def mwf_metric(self, window_selected_temp):
- """
- Find multi_window_finder method metric value for a chosen window size
+ """ Find multi_window_finder method metric value for a chosen window size
 
- :param window_selected_temp: temporary window selected
- :return: value which is the MWF distance metric
+ Arg:
+ window_selected_temp: temporary window selected
+
+ Return:
+ value which is the MWF distance metric
  """
  coeff_temp = 1 / window_selected_temp
  m_values = []
@@ -193,12 +209,15 @@ def mwf_metric(self, window_selected_temp):
  distance_k = sum(np.log10(abs(m_values - np.mean(m_values))))
  return distance_k
 
- def top_local_minimum(self, distance_scores):
- """
- Find a list of local minimum for multi_window_finder method
+ @staticmethod
+ def top_local_minimum(distance_scores):
+ """ Find a list of local minimum for multi_window_finder method
+
+ Arg:
+ distance_scores: list of distance scores from mwf_metric
 
- :param distance_scores: list of distance scores from mwf_metric
- :return: list of index where narray has minimum, max id for distance_scores list
+ Return:
+   list of index where narray has minimum, max id for distance_scores list
  """
  id_max = distance_scores.index(max(distance_scores))
  score_temp = distance_scores[id_max:]
@@ -210,6 +229,11 @@ def top_local_minimum(self, distance_scores):
  return id_local_minimum_list, id_max
 
  def runner_wss(self):
+ """ Initial function to optimize hyperparameter.
+
+ Returns:
+ selected parameter and scores list
+ """
  if int(len(self.time_series)) <= self.window_min:
  window_size_selected, list_score = int(len(self.time_series)), []
  else: