cjk-data-profiling

cjk-data-profiling is a context-manager package for generating ydata-profiling / fg-data-profiling reports with Japanese, Chinese, or Korean text. It temporarily configures CJK fonts for Matplotlib and WordCloud and prevents report interaction DOM-ID collisions caused by column names.

Installation

pip install cjk-data-profiling

Install the optional profiling dependency when needed:

pip install "cjk-data-profiling[profiling]"

Usage

Wrap report generation in cjk_data_profiling.

import pandas as pd
from data_profiling import ProfileReport

from cjk_data_profiling import cjk_data_profiling

dataframe = pd.DataFrame({"city": ["東京", "大阪"]})

with cjk_data_profiling("ja"):
    report = ProfileReport(dataframe)
    report.to_file("report.html")

Language-specific values are "ja", "zh-cn", "zh-tw", and "ko". Use the additional "mixed" value only when a Noto Sans CJK font is installed; it requires one font that covers Japanese, Chinese, and Korean text.

To prefer a different installed font, replace the candidate list for the relevant language before entering the context. The selected font must include the CJK characters used by the report.

from cjk_data_profiling import FONT_CANDIDATES, cjk_data_profiling

FONT_CANDIDATES["ja"] = ["Yu Gothic", "Meiryo"]

with cjk_data_profiling("ja"):
    ...

All compatibility patches are enabled by default. To opt out of one, pass its keyword-only switch. For example, leave WordCloud's default font unchanged:

with cjk_data_profiling("ja", enable_wordcloud=False):
    report = ProfileReport(dataframe)

The available switches are enable_matplotlib, enable_wordcloud, and enable_slugify.

The context temporarily changes process-wide Matplotlib and WordCloud defaults, as well as the installed profiling package's report-structure DOM-ID helper. Concurrent report generation from multiple threads in the same process is not supported. Both this package and fg-data-profiling temporarily modify Matplotlib global state while rendering. Do not make this context active in overlapping threads, including with different languages. If reports must be generated concurrently, use separate processes. For threaded applications, serialize the complete report-generation operation with an application-level lock instead.

Testing

pytest
pytest -m integration

Run the non-integration suite on every supported Python version locally:

uv sync --group dev
uv run tox

tox obtains Python 3.11, 3.12, 3.13, and 3.14 through uv, then runs pytest -m "not integration". Report-generation integration tests remain opt-in. The interaction-selector integration test writes *-interaction-selectors.html files to test_output/integration-reports/. Open one in a browser to manually confirm that every Interaction selector, including "a b", "a-b", and CJK column names, displays its matching plot.

Integration reports and WordCloud images are written to test_output/integration-reports/ and are ignored by Git.