compute-variability.py: Enhance peak detection for 2-peaks-linear algorithm

Change-Id: I6099890fb71730d7f48782f7ead27a0ca08ec77e

1 files changed, 10 insertions, 5 deletions

diff --git a/compute-variability.py b/compute-variability.py
index d655142..2bf9a02 100755
--- a/compute-variability.py
+++ b/compute-variability.py
@@ -6,6 +6,7 @@ import pandas as pd
 import argparse
 import numpy as np
 import math
+from scipy.signal import find_peaks
 
 import metric_utils
 
@@ -38,17 +39,22 @@ Routine to compute weighted average. """
 def compute_weighted_total(values, aggregate_method, aggregate_weights):
 
     v = values.iloc[2:-1].astype("float").to_numpy()
+    # DEBUG print("v        = ", v)
 
     if aggregate_weights=="linear":
         w = np.linspace(100, 0, len(v))
     elif aggregate_weights=="flat":
         w = np.linspace(100, 100, len(v))
     elif aggregate_weights=="2-peaks-linear":
-        indexes=[i for i, x in enumerate(v) if x >= 0]
+        values_interp = pd.Series(v).interpolate()
+        # DEBUG print("v_interp = ", values_interp.to_numpy())
+        peaks = find_peaks(values_interp, height=0)
+        # DEBUG print("peaks = ", peaks)
         idx=len(v)
-        if len(indexes)>=2:
-            idx=indexes[1]+1
+        if len(peaks[0])>=2:
+            idx=peaks[0][1]
         w = [*np.linspace(100, 50, idx) , *np.linspace(0, 0, len(v)-idx)]
+        # DEBUG print("w = ", w)
     else:
         raise AssertionError("weights option should be either 'flat, 'linear' or '2-peaks-linear'")
 
@@ -71,8 +77,6 @@ def compute_weighted_total(values, aggregate_method, aggregate_weights):
         raise AssertionError("method option should be either 'avg' or 'max'")
 
     # DEBUG
-    # print("    ", masked_values)
-    # print("  x ", w)
     # print("  = ", round(total,2))
     return round(total,2)
 
@@ -140,6 +144,7 @@ def compute_threshold(metric, aggregate_method, aggregate_weights):
             history_df = pd.merge(compare_df, history_df, how="outer", on=["benchmark", "symbol"])
 
     for i, row in history_df.iterrows():
+        # DEBUG print("\n============ ", row["benchmark"], row["symbol"])
         total = compute_weighted_total(history_df.loc[i], aggregate_method, aggregate_weights)
         history_df.loc[i, metric+'_variation_total'] = total