feat: Add pipeline for running time (#54)

feat: Add pipeline for rt

examples/pipeline.py → examples/pipeline/dwell_time.py

@@ -25,7 +25,7 @@ def datetime_from_string(datetime_string: str) -> datetime:
     )
     incremental_learning = True
     is_buffer_enabled = False
-    folder_path_to_save_result = "../experiments"
+    folder_path_to_save_result = "../../experiments"
     experiment_label = "m-xgb-s-xgb_model"
 
     run_dt_exp(

examples/pipeline/running_time.py

@@ -0,0 +1,51 @@
+from datetime import datetime
+
+from ibat.concept_drift_detector.strategies import DDM
+from ibat.datasets import BUS_654_FEATURES_ADDED_RUNNING_TIMES
+from ibat.pipeline import run_rt_exp
+
+
+def datetime_from_string(datetime_string: str) -> datetime:
+    return datetime.strptime(datetime_string, "%Y-%m-%d")
+
+
+if __name__ == "__main__":
+    running_time_df = BUS_654_FEATURES_ADDED_RUNNING_TIMES.dataframe
+    running_time_df = running_time_df.dropna(subset=["start_time"])
+    running_time_df = running_time_df[running_time_df["direction"] == 1]
+    running_time_df[["direction", "segment"]] = running_time_df[
+        ["direction", "segment"]
+    ].astype(int)
+
+    historical_data_starting_from = datetime_from_string("2021-10-01")
+    historical_data_ending_at = datetime_from_string("2022-02-01")
+    streaming_data_starting_from = datetime_from_string("2022-02-01")
+    streaming_data_ending_at = datetime_from_string("2022-11-01")
+    time_interval = 60 * 2
+    chunk_size = 100
+    active_strategy = True
+    cdd_strategy = DDM(
+        warning_level=0.1,
+        drift_level=1.5,
+        min_num_instances=1,
+    )
+    incremental_learning = True
+    is_buffer_enabled = False
+    folder_path_to_save_result = "../../experiments"
+    experiment_label = "m-xgb-s-xgb_model"
+
+    run_rt_exp(
+        rt_df=running_time_df,
+        hist_start=historical_data_starting_from,
+        hist_end=historical_data_ending_at,
+        stream_start=streaming_data_starting_from,
+        stream_end=streaming_data_ending_at,
+        interval_min=time_interval,
+        chunk_size=chunk_size,
+        active_strategy=active_strategy,
+        cdd_strategy=cdd_strategy,
+        incremental_learning=incremental_learning,
+        is_buffer_enabled=is_buffer_enabled,
+        output_parent_dir=folder_path_to_save_result,
+        label=experiment_label,
+    )

experiments/ex-m-xgb-s-xgb_model-2024-05-26_15-38-31/README.md

@@ -0,0 +1,35 @@
+
+# Experiment: m-xgb-s-xgb_model
+
+## Parameters
+- Historical data starting from: 2021-10-01 00:00:00
+- Historical data ending at: 2022-02-01 00:00:00
+- Streaming data starting from: 2022-02-01 00:00:00
+- Streaming data ending at: 2022-11-01 00:00:00
+- Minimum waiting time: 120 minutes
+- Minimum required amount of data: 100
+- Batch processing technique: HYBRID
+- Concept drift handling strategy: ACTIVE
+- Concept drift detection algorithm: DDM
+- Strategy to update the outdated model: INCREMENTAL LEARNING
+- Is buffer enabled: False
+- DDM:
+
+| Attribute | Value |
+|---|---|
+| Warning level factor | 0.1 |
+| Drift level factor | 1.5 |
+| Minimum numbers of instances/batches to start looking for changes | 1 |
+
+
+## Results
+- Model performance metrics:
+
+| Model                                     | MAE (s)              | RMSE (s)              | MAPE (%)              |
+|-------------------------------------------|----------------------|-----------------------|-----------------------|
+| Base model without concept drift handling | 38.827 | 63.670 | 0.284 |
+| Base model with concept drift handling    | 36.461      | 59.327      | 0.242      |
+
+- Error reduction percentage in terms of MAE: 6.094 %
+- Average processing time after the batch preparation: 25.202 ms
+

ibat/pipeline/__init__.py

@@ -0,0 +1,2 @@
+from .dwell_time import run_dt_exp
+from .running_time import run_rt_exp