csvmedic 0.1.3

Automatic locale-aware CSV and Excel reader with encoding, delimiter, date format, and number locale detection.

csvmedic

One-line CSV reading that auto-detects encoding, delimiter, dates, number formats, and booleans—so you don’t lose leading zeros or misinterpret 03/04/2025.

import csvmedic

df = csvmedic.read("export.csv")
print(df.diagnosis)  # What was detected and converted

Features

Detects	Examples
Encoding	UTF-8, Windows-1252, ISO-8859-1, Shift-JIS, BOM
Delimiter	Comma, semicolon, tab, pipe
Dates	DD-MM vs MM-DD resolved from data; ISO, European, US formats
Numbers	European (1.234,56) vs US (1,234.56)
Booleans	Yes/No, Ja/Nein, Oui/Non, Sí/No, and more
Strings	Preserves leading zeros (e.g. IDs 00742)

All decisions are recorded on the returned DataFrame’s .diagnosis attribute.

Installation

pip install csvmedic

Optional extras:

• csvmedic[fast] — better dialect detection (clevercsv)
• csvmedic[excel] — .xlsx support (openpyxl)
• csvmedic[all] — both

Usage

Override detection when you know better:

df = csvmedic.read(
    "file.csv",
    encoding="utf-8",
    delimiter=";",
    dayfirst=True,
    preserve_strings=["ID"],
    confidence_threshold=0.75,
)

Inspect without converting:

profile = csvmedic.read_raw("file.csv")
print(profile.summary())

Schema pinning (recurring files):

df = csvmedic.read("monthly_export.csv")
csvmedic.save_schema(df.attrs["diagnosis"].file_profile, "monthly_export.csvmedic.json")
# Next run: skip detection
df2 = csvmedic.read("monthly_export.csv", schema="monthly_export.csvmedic.json")

Batch with consensus:

dfs = csvmedic.read_batch(["jan.csv", "feb.csv", "mar.csv"], use_consensus=True)

Compare with pandas:

result = csvmedic.diff("file.csv")
print(result.summary())
# result.pandas_df, result.csvmedic_df, result.sample_differences

Date disambiguation

For ambiguous values like 03/04/2025, csvmedic uses the column: if any value has day > 12 (e.g. 25/03/2025), the column is treated as day-first. It also uses cross-column inference, separator hints (e.g. period → European), and order. If still ambiguous, the column stays as string and is marked in the diagnosis.

Documentation

• Quickstart
• How it works
• API reference
• FAQ

License

MIT