preliminary edits

.gitignore

@@ -8,6 +8,5 @@ docs/build/
 .local/
 cover/
 *.pickle
-.ipynb_checkpoints
 */.ipynb_checkpoints/*
 mlfinlab.egg-info/*

mlfinlab/online_portfolio_selection/pattern_matching/correlation_driven_nonparametric_learning.py

@@ -21,7 +21,9 @@ def __init__(self, window, rho):
         Initializes Correlation Driven Nonparametric Learning with the given window and rho value.
 
         :param window: (int) Number of windows to look back for similarity sets.
-        :param rho: (float) Threshold for similarity.
+        :param rho: (float) Threshold for similarity. Rho should set in the range of [-1, 1].
+                    Lower rho values will classify more periods as being similar, and higher values
+                    will be more strict on identifying a period as similarly correlated.
         """
         self.window = window
         self.rho = rho
@@ -33,7 +35,7 @@ def _initialize(self, asset_prices, weights, resample_by):
         Initializes the important variables for the object.
 
         :param asset_prices: (pd.DataFrame) Historical asset prices.
-        :param weights: (list/np.array/pd.Dataframe) Initial weights set by the user.
+        :param weights: (list/np.array/pd.DataFrame) Initial weights set by the user.
         :param resample_by: (str) Specifies how to resample the prices.
         """
         super(CorrelationDrivenNonparametricLearning, self)._initialize(asset_prices, weights,
@@ -60,8 +62,10 @@ def _update_weight(self, time):
         """
         # Create similar set.
         similar_set = []
+
         # Default is uniform weights.
         new_weights = self._uniform_weight()
+
         # Calculate for similar sets if time is greater or equal to window size.
         if time >= self.window:
             # Iterate through past windows.
@@ -70,6 +74,7 @@ def _update_weight(self, time):
                 if self.corr_coef[time - self.window + 1][past_time] > self.rho:
                     # Append the time for similar set.
                     similar_set.append(past_time + self.window)
+
             if similar_set:
                 # Choose the corresponding relative return periods.
                 optimize_array = self.relative_return[similar_set]
@@ -113,11 +118,14 @@ def calculate_rolling_correlation_coefficient(self):
         """
         # Flatten the array.
         flattened = self.relative_return.flatten()
+
         # Set index of rolled window.
         idx = np.arange(self.number_of_assets * self.window)[None, :] + self.number_of_assets * \
               np.arange(self.length_of_time - self.window + 1)[:, None]
+
         # Retrieve the results of the rolled window.
         rolled_returns = flattened[idx]
+
         # Calculate correlation coefficient.
         rolling_corr_coef = np.nan_to_num(np.corrcoef(rolled_returns), nan=0)
         return rolling_corr_coef

mlfinlab/online_portfolio_selection/pattern_matching/correlation_driven_nonparametric_learning_k.py

@@ -25,7 +25,7 @@ def __init__(self, window, rho, k):
         windows, rho values, and k experts.
 
         :param window: (int) Number of windows to look back for similarity sets. (1, 2, ..., w).
-        :param rho: (float) Number of rho values for threshold. (0, 1/rho, ..., (rho-1)/rho).
+        :param rho: (int) Number of rho values for threshold. (0, 1/rho, ..., (rho-1)/rho).
         :param k: (int) Number of top-k experts.
         """
         self.window = window
@@ -85,11 +85,11 @@ def _generate_experts(self):
         pointer = 0
 
         # Window from 1 to self.window.
-        for n_window in range(self.window):
+        for n_window in range(1, self.window):
             # Rho from 0 to (rho - 1)/rho.
             for n_rho in range(self.rho):
                 # Assign experts with parameters (n_window + 1, n_rho/rho).
-                self.expert_params[pointer] = [n_window + 1, n_rho/self.rho]
+                self.expert_params[pointer] = [n_window, n_rho/self.rho]
                 # Next pointer.
                 pointer += 1
         # Assign parameters.