Fixed a bug affecting fictional confidence regions, added smoothing o…

…ption to baseline, improved visualization
AutoTuningAssociation · Nov 7, 2023 · 401dd3c · 401dd3c
1 parent 331762c
commit 401dd3c
Show file tree

Hide file tree

Showing 4 changed files with 14 additions and 12 deletions.
diff --git a/experiment_files/example_visualizations.json b/experiment_files/example_visualizations.json
@@ -50,11 +50,6 @@
         ]
     },
     "strategies": [
-        {
-            "name": "greedy_mls",
-            "strategy": "greedy_mls",
-            "display_name": "Greedy Multi-Start Local Search"
-        },
         {
             "name": "greedy_ils",
             "strategy": "greedy_ils",

diff --git a/src/autotuning_methodology/baseline.py b/src/autotuning_methodology/baseline.py
@@ -9,7 +9,7 @@
 import numpy as np
 from scipy.interpolate import PchipInterpolator
 
-from autotuning_methodology.curves import CurveBasis, get_indices_in_array
+from autotuning_methodology.curves import CurveBasis, get_indices_in_array, moving_average
 from autotuning_methodology.searchspace_statistics import SearchspaceStatistics
 
 
@@ -170,7 +170,13 @@ def get_curve_over_fevals(  # noqa: D102
 
     def get_curve_over_time(self, time_range: np.ndarray, dist=None, confidence_level=None) -> np.ndarray:  # noqa: D102
         fevals_range = self.time_to_fevals(time_range)
-        return self.get_curve_over_fevals(fevals_range, dist, confidence_level)
+        curve_over_time = self.get_curve_over_fevals(fevals_range, dist, confidence_level)
+        smoothing_factor = 0.0
+        if smoothing_factor > 0.0:
+            window_size = min(time_range.size, ceil(time_range.size * smoothing_factor))
+            if time_range.size > 1 and window_size < 1:
+                curve_over_time = moving_average(curve_over_time, window_size)
+        return curve_over_time
 
     def get_standardised_curve(  # noqa: D102
         self, range: np.ndarray, strategy_curve: np.ndarray, x_type: str

diff --git a/src/autotuning_methodology/curves.py b/src/autotuning_methodology/curves.py
@@ -402,7 +402,7 @@ def _get_curve_split_real_fictional_parts(
         curve: np.ndarray,
         curve_lower_err: np.ndarray,
         curve_upper_err: np.ndarray,
-        smoothing_factor=0.005,
+        smoothing_factor=0.0,
     ) -> tuple[int, np.ndarray, np.ndarray, np.ndarray, np.ndarray, np.ndarray, np.ndarray, np.ndarray, np.ndarray]:
         """Split the provided curves based on the real_stopping_point_index.
 

diff --git a/src/autotuning_methodology/visualize_experiments.py b/src/autotuning_methodology/visualize_experiments.py
@@ -723,7 +723,7 @@ def normalize_multiple(curves: list) -> tuple:
                         color=color,
                         ls="dashed",
                     )
-                ax.plot(x_axis_range_fictional, curve_fictional, color=color, ls="dashed")
+                ax.plot(x_axis_range_fictional, curve_fictional, color=color, alpha=0.8, ls="dashed")
 
         # # plot cutoff point
         # def plot_cutoff_point(cutoff_percentiles: np.ndarray, show_label=True):
@@ -904,12 +904,12 @@ def plot_strategies_aggregated(
                 )
                 if (
                     real_stopping_point_index < time_range.shape[0]
-                    and real_stopping_point_index < len(strategies_lower_err) - 1
+                    and real_stopping_point_index < len(strategy_lower_err) - 1
                 ):
                     ax.fill_between(
                         time_range[real_stopping_point_index:],
-                        strategies_lower_err[real_stopping_point_index:],
-                        strategies_upper_err[real_stopping_point_index:],
+                        strategy_lower_err[real_stopping_point_index:],
+                        strategy_upper_err[real_stopping_point_index:],
                         alpha=0.15,
                         antialiased=True,
                         color=color,
@@ -928,6 +928,7 @@ def plot_strategies_aggregated(
                     time_range[real_stopping_point_index:],
                     strategy_performance[real_stopping_point_index:],
                     color=color,
+                    alpha=0.8,
                     ls="dashed",
                 )
             print(f" | performance of {displayname}: {round(np.mean(strategy_performance), 3)}")