Repository navigation
to_fills_dataframe() drops the feed timezone, so fills cannot be joined back to the feed #104
Description
Version: ml4t-backtest 0.1.8
With a feed whose timestamp is Datetime("us", "UTC"), Strategy.on_data receives tz-aware timestamps, but result.to_fills_dataframe()["timestamp"] is naive Datetime("us"). Joining fills to the feed, or to anything a strategy logged from on_data, raises:
polars.exceptions.SchemaError: datatypes of join keys don't match - `timestamp`: datetime[μs] on left does not match `timestamp`: datetime[μs, UTC] on right
Repro: a one-asset UTC feed of five daily bars, a strategy that submits one order, then
fills = result.to_fills_dataframe()
fills.schema["timestamp"] # Datetime(time_unit='us', time_zone=None)
fills.join(feed_df.select("timestamp", "asset"), on=["timestamp", "asset"]) # SchemaErrorExpected: result frames keep the feed's time zone, as on_data does. Worth checking the other to_*_dataframe() exports for the same loss.
Activity
fix(results): preserve feed timezone in exported records
stefan-jansen commentedon Sep 24, 2026
Verified on PR #119 at commit 8343029. The focused tests in tests/test_result.py passed locally (81 passed); the remote coverage job passed. Exported result records preserve the feed timestamp timezone through artifact round trips.
Fix Backtest correctness issues and build guided documentation