feature: add pattern detection

pipit/trace.py

@@ -802,3 +802,62 @@ def multirun_analysis(
         combined_df = combined_df[function_sums.sort_values(ascending=False).index]
 
         return combined_df
+
+    def detect_pattern(
+        self,
+        start_event,
+        iterations=None,
+        window_size=None,
+        process=0,
+        metric="time.exc",
+    ):
+        import stumpy
+
+        enter_events = self.events[
+            (self.events["Name"] == start_event)
+            & (self.events["Event Type"] == "Enter")
+            & (self.events["Process"] == process)
+        ]
+
+        leave_events = self.events[
+            (self.events["Name"] == start_event)
+            & (self.events["Event Type"] == "Leave")
+            & (self.events["Process"] == process)
+        ]
+
+        # count the number of enter events to
+        # determine the number of iterations if it's not
+        # given by the user.
+        if iterations is None:
+            iterations = len(enter_events)
+
+        # get the first enter and last leave of
+        # the given event. we will only investigate
+        # this portion of the data.
+        first_loop_enter = enter_events.index[0]
+        last_loop_leave = leave_events.index[-1]
+
+        df = self.events.iloc[first_loop_enter + 1 : last_loop_leave]
+        filtered_df = df.loc[(df[metric].notnull()) & (df["Process"] == process)]
+        y = filtered_df[metric].values[:]
+
+        if window_size is None:
+            window_size = int(len(y) / iterations)
+
+        matrix_profile = stumpy.stump(y, window_size)
+        dists, indices = stumpy.motifs(y, matrix_profile[:, 0], max_matches=iterations)
+
+        # Gets the corresponding portion from the original
+        # dataframe for each pattern.
+        patterns = []
+        for idx in indices[0]:
+            end_idx = idx + window_size
+
+            match_original = self.events.loc[
+                self.events["Timestamp (ns)"].isin(
+                    filtered_df.iloc[idx:end_idx]["Timestamp (ns)"].values
+                )
+            ]
+            patterns.append(match_original)
+
+        return patterns