feat(Events): added EventFilter, EventLabeller (#186)

labeling/labellers/fixed_time_three_class_imbalanced.py

@@ -0,0 +1,44 @@
+from ..types import EventLabeller, EventsDataFrame, ForwardReturnSeries
+import pandas as pd
+
+class FixedTimeHorionThreeClassImbalancedEventLabeller(EventLabeller):
+
+    time_horizon: int
+
+    def __init__(self, time_horizon: int = 1):
+        self.time_horizon = time_horizon
+
+    def label_events(self, event_start_times: pd.DatetimeIndex, forward_returns: ForwardReturnSeries) -> EventsDataFrame:
+
+        event_candidates = forward_returns[event_start_times]
+
+        def get_bins_threeway(x):
+            bins = pd.qcut(event_candidates, 4, retbins=True, duplicates = 'drop')[1]
+
+            if len(bins) != 5:
+                # if we don't have enough data for the quantiles, we'll need to add hard-coded values
+                lower_bound = bins[0]
+                upper_bound = bins[-1]
+                bins = [lower_bound] + [-0.02, 0.0, 0.02] + [upper_bound]
+            return bins
+        bins = get_bins_threeway(event_candidates)
+
+        def map_class_threeway(current_value):
+            lower_threshold = bins[1]
+            upper_threshold = bins[3]
+            if current_value <= lower_threshold:
+                return -1
+            elif current_value > lower_threshold and current_value < upper_threshold:
+                return 0
+            else:
+                return 1
+        labels = event_candidates.map(map_class_threeway)
+        
+        return pd.DataFrame({
+            'start': event_start_times,
+            'end': event_start_times + pd.Timedelta(days=self.time_horizon),
+            'label': labels,
+            'returns': forward_returns[event_start_times]
+        })
+
+