CSV to Parquet Converter

Parquet schema: explicit strings and empty cells

Values such as 00123, 2026-09-13 and true are written as strings, not numbers, dates or Booleans. Empty cells stay empty strings. Explicit typing preserves identifiers and original spelling, but numerical analysis requires a deliberate cast. This converter does not infer nested structures, decimal precision or timestamp time zones.

Read the converted Parquet file in Python

import pandas as pd

# Requires a Parquet engine such as pyarrow.
df = pd.read_parquet("converted.parquet")
print(df.dtypes)
print(df.head())
# Convert a known numeric column explicitly:
# df["amount"] = pd.to_numeric(df["amount"], errors="raise")

Inspect the file with DuckDB

SELECT * FROM read_parquet('converted.parquet') LIMIT 10;
DESCRIBE SELECT * FROM read_parquet('converted.parquet');

Replace the example filename with your actual download name. Compare row counts, column names and leading-zero values with the source. A downstream schema that requires integer or date columns may reject these string columns until you cast them explicitly.

Snappy compression and large CSV limits

Exports use Snappy compression. Compression ratios depend on the data, and metadata can outweigh savings for small tables. The browser retains the parsed table, creates column arrays and builds an output buffer. For recurring large-dataset jobs, use a local Python or DuckDB workflow. Incomplete input is blocked rather than silently exporting a partial dataset.

Export a JSON array instead · CSV → JSONL