bankcsv

Load a folder of a bank's CSV statement exports into a single, normalised pandas.DataFrame. Overlapping downloads (the bank only lets you export a rolling window, so you end up with many overlapping files) are reconciled with a date-range replacement strategy: the most recent export wins for every date it covers.

Supported banks: Bankwest, ANZ, Macquarie, CBA.

Installation

pip install -e ".[dev]"

Usage

from bankcsv import Banking

b = Banking()
df = b.bankwest("~/iCloudDrive/Bankwest")   # folder known to be Bankwest
df = b.anz("~/Downloads")                    # folder known to be ANZ
df = b.macquarie("~/Downloads/macquarie")
df = b.cba("~/Downloads/cba")

Auto-detection

ingest walks every CSV under the given path(s) — a folder, a single .csv, or an iterable of either — and offers each file to every bank loader in turn. The first loader that parses the whole file cleanly claims it; files nothing recognises are logged and skipped. Claimed files are merged per bank and concatenated, date-sorted:

df = b.ingest(["~/iCloudDrive/Bankwest", "~/Downloads"])
df.groupby("bank").size()

Detection is by content, not filename. Loaders are tried most-specific first:

All frames use the same columns (see below), so pd.concat([...]) across banks just works. The load_bankwest / load_anz / load_macquarie / load_cba / ingest functions are also exported for direct use.

Transaction schema

Every loader (load_bankwest, load_anz, …) returns a pandas.DataFrame with the same nine columns, in this order. The columns are always present (mandatory). Required fields are additionally guaranteed to be non-null on every row; optional fields are None when the source does not provide them.

Money is normalised to a single signed amount (decimal.Decimal): credits are positive, debits negative. date is a datetime.date.

The frame is object-dtype throughout and gaps are Python None, never NaN, so Decimal columns stay exact and x is None works. An optional column that is None on every row means that bank does not expose it; call df.dropna(axis=1, how="all") for a per-bank trimmed view.

Source mapping

Headerless banks are shown by column index; — means the export does not carry that field (so the column is None for every row of that bank).

Not carried: ANZ col[5]/col[7] (always empty); Bankwest Cheque; Macquarie Category / Subcategory / Original Description (used only for detection).

Development

pip install -e ".[dev]"
pytest

Licence

MIT — see LICENSE.