Add pl.DataFrame.to_markdown

[mkleinbort-ic]

Description

It'd be good to have a to_markdown method at the dataframe level.

The string repr is almost right, just needs some tweeking to be valid markdown.

A very bad implementation in Python would be:

def to_markdown(df:pl.DataFrame)->str:
      '''Returns the repr of the dataframe as valid markdown'''

      lines = (df
            .pipe(str)
            .replace('┆','|')
            .replace('╪','|')
            .replace('╡','|')
            .replace('╞','|')
            .replace(' --- ', '     ')
            .replace('═', '-')
            .replace('┌', '|')
            .replace('┐', '|')
            .replace('└', '|')
            .replace('┘', '|')
            .replace('─', '')
            .replace('┬', '')
            .replace('┴', '')
            .replace('||', '')
            .split('\n')
      )

      lines_to_show = [lines[2]]+lines[5:] # column names + values (skips shape & data type)
      
      ans = '\n'.join([l.strip('│ ').strip('|') for l in lines_to_show])

      return ans

[baggiponte]

I also received a feature request to cast to HTML; if one is accepted then the other would be pretty much straighforward.

[deanm0000]

doesn't _repr_html already cast it to html or am I misunderstanding something?

[cjackal]

Another (less-effort) implementation:

import polars as pl
import io

def to_markdown(df: pl.DataFrame) -> str:
    buf = io.StringIO()
    with pl.Config(
        tbl_formatting="ASCII_MARKDOWN",
        tbl_hide_column_data_types=True,
        tbl_hide_dataframe_shape=True,
    ):
        print(df, file=buf)
    buf.seek(0)
    return buf.read()

One may need more config options like tbl_rows=-1 though.

[MarcoGorelli]

pandas does this by deferring to tabulate - there's a feature there about adding support for polars astanin/python-tabulate#258
I guess contributing Polars support to tabulate would be the easiest way to make it happen

[cjackal]

@MarcoGorelli I thought the opposite trying to implement pyarrow table support to it a while ago; the way tabulate support pandas is a bit too rough (checking hasattr(df, "value"), hasattr(df, "index"), hasattr(df.index, "name"), ..., ) for other third party dataframe libraries to support it without first converting to pandas.

[alexander-beedie]

Another (less-effort) implementation:

Nice one; I think it's relatively unknown, but the Config object can also function as a decorator¹, which works well with that pattern, eg:

@pl.Config( 
    tbl_formatting = "ASCII_MARKDOWN",        
    tbl_hide_column_data_types = True,
    tbl_hide_dataframe_shape = True,
)
def frame_to_markdown( df: pl.DataFrame ) -> str:
    return str(df)

If we could add left/right/center align options to the markdown output we'd have the basics covered at least 🤔

Footnotes

1. https://github.com/pola-rs/polars/pull/9307 ↩