THULab/traffic_prediction_dataset_engineered

Traffic Prediction Engineered Dataset (TsFile)

Apache TsFile version of jojonocode/traffic-prediction-dataset-engineered.

Overview

Hourly vehicle counts at four key junctions of an urban road network, published with the feature engineering used by the author's LSTM traffic-prediction experiments. Besides the raw hourly count, every row carries the project's derived features: calendar fields, cyclic hour encodings (hour_sin / hour_cos) and lagged vehicle counts (veh_lag_1/2/3/24).

• Tables: traffic_test, traffic_train
• Rows: 48,096 across 2 TsFile table(s).
  • traffic_test: 9,621 rows, 2017-03-01 09:00:00 → 2017-06-30 23:00:00 (UTC)
  • traffic_train: 38,475 rows, 2015-11-01 00:00:00 → 2017-05-25 18:00:00 (UTC)

Schema (TsFile structure)

traffic_test.tsfile

• Time (INT64, milliseconds) — the source DateTime timestamp as INT64 milliseconds.
• TAG: Junction — device dimension(s); query one device with WHERE Junction=....
• FIELD (12): Vehicles (int64), ID (int64), hour (int64), dayofweek (int64), month (int64), is_weekend (int64), hour_sin (double), hour_cos (double), veh_lag_1 (double), veh_lag_2 (double), veh_lag_3 (double), veh_lag_24 (double)

traffic_train.tsfile

• Time (INT64, milliseconds) — the source DateTime timestamp as INT64 milliseconds.
• TAG: Junction — device dimension(s); query one device with WHERE Junction=....
• FIELD (12): Vehicles (int64), ID (int64), hour (int64), dayofweek (int64), month (int64), is_weekend (int64), hour_sin (double), hour_cos (double), veh_lag_1 (double), veh_lag_2 (double), veh_lag_3 (double), veh_lag_24 (double)

Conversion notes

• The source ships a train/test split; each split is converted into its own TsFile (traffic_train.tsfile, traffic_test.tsfile) and the split membership is preserved.
• Junction (1-4) is declared as the TAG (device) column: query one junction with WHERE Junction=1.
• Time is the source DateTime (hourly, local wall-clock) as INT64 milliseconds.
• Not converted: traffic_engineered_full.csv — verified row-for-row identical to the concatenation of splits/train.csv and splits/test.csv (48,096 rows each, same columns and values).
• Not converted: traffic_original.csv — the pre-feature-engineering copy of the same 48,120 hourly junction records; the engineered train/test files carry the same DateTime/Junction/Vehicles values plus the derived features.

Usage

Install the Apache TsFile Python SDK (pip install tsfile) and read a converted file:

from pathlib import Path
from tsfile import TsFileReader

path = Path("traffic_test.tsfile")
with TsFileReader(str(path)) as reader:
    schemas = reader.get_all_table_schemas()
    print("tables:", list(schemas))
    table_name = next(iter(schemas))
    table = schemas[table_name]
    columns = [column.get_column_name() for column in table.get_columns()]
    print("columns:", columns)
    field_names = [
        column.get_column_name()
        for column in table.get_columns()
        if column.get_column_name() not in {"Time", "time"}
    ]
    if field_names:
        with reader.query_table(table_name, field_names[:3], batch_size=1024) as result:
            batch = result.read_arrow_batch()
            if batch is not None:
                print(batch.to_pandas().head())

Source & license

• Original dataset: jojonocode/traffic-prediction-dataset-engineered
• Source card title / publisher: Traffic Prediction Engineered Dataset
• License: cc-by-4.0 (as declared by the original dataset).