lastra 0.8.3

Python reader/writer for the Lastra columnar time series format.
Bit-exact compatibility with lastra-java and lastra-ts.

Status

Reader and writer feature-complete, aligned with lastra-java. Wire-format spec lives in FORMAT.md. See the PyPI badge above for the latest version.

Install

pip install lastra

Usage

from lastra import LastraReader

with open("ohlcv.lastra", "rb") as f:
    r = LastraReader.from_stream(f)

    # Read only what you need — other columns are not decompressed
    ts = r.read_series_long("ts")        # numpy int64 array
    close = r.read_series_double("close") # numpy float64 array

    # Column metadata
    meta = r.get_series_column("ema1").metadata
    # {"indicator": "ema", "periods": "10"}

    # Events (independent timestamps)
    sig_ts = r.read_event_long("ts")
    sig_data = r.read_event_binary("data")  # list[bytes | None]

Pandas / Polars / Arrow adapters (planned for 0.9):

df = LastraReader(f).to_pandas()      # all series columns
pl_df = LastraReader(f).to_polars()
table = LastraReader(f).to_arrow()

Why Lastra

Lastra is a columnar file format optimised for numeric time series — financial tick data, IoT sensors, infrastructure metrics. It applies semantic compression per column:

• ALP: decimal-aware → ~3-4 bits/value for prices at 2 decimal places
• Pongo: decimal erasure + Gorilla XOR → ~18 bits/value
• Gorilla: XOR for volatile metrics
• DELTA_VARINT: ~1 byte/value for regular timestamps

A typical OHLCV column at 2 decimal places compresses to ~13× the raw size — about 2× better than Apache Parquet (SNAPPY) or ORC for the same data. See the lastra-java README for benchmarks.

Reference implementations

Conversion to/from Parquet, CSV, Arrow

See QTSurfer/lastra-convert-py — Python CLI port of the QTSurfer/lastra-convert (Java) tool.

License

Copyright 2026 Wualabs LTD. Apache License 2.0 — see LICENSE.