Incorrect Precision Repr

[mck-star-yar]

I am trying to align the precision representation for the column and it doesn't work. Assume the following exmaple:

from itables import show

df = pd.Series([1.0, 1.123])
with pd.option_context("display.float_format", "{:.3f}".format):
    show(df)

It shows

0
1
1.123

But expected

0
1 .000
1.123

The issue seems to be in itables/datatables_format.py:: _format_column.
After formatted input is cast to numpy array, it looses the formatting:

    if dtype_kind == "f":
        try:
            return np.array(x).astype(float)
        except ValueError:
            pass

removing this line preserves proper formatting.

itables version == '1.4.4'

[mwouts]

Hey @mck-star-yar , thanks for reporting this.

You're right, we convert formatted numbers back to numbers. The motive for doing so is to get the sorting right in datatables.

One option would be to do the formatting in Javascript, but I am not so sure this will really work. For instance, this example uses $ (i.e. jQuery), which is a global Javascript variable that we don't provide anymore. Let us know if you find a better way to do that on the Javascript side!

Maybe a more pragmatic approach would be an option floats_as_strings=False by default, both in itables.options and show, and you could set it to True to discard the conversion back to float?

[mwouts]

Actually the JS example might work! At least I see that we have a similar example in the docs and that one works: https://mwouts.github.io/itables/advanced_parameters.html#advanced-cell-formatting-with-js-callbacks

[mck-star-yar]

Unfortunately, I'm not writing in JS. Can you please provide a code snippet for this JS function?

I tried following the examples you shared, but those doesn't make much sense to me. Tried like this but it fails:

from itables import show, JavascriptFunction
import pandas as pd


df = pd.Series([1.0, 1.123])
with pd.option_context("display.float_format", "{:.3f}".format):
    show(
        df,
        columnDefs=[
            {"targets": 0, "render": "DataTable.render.number(',', '.', 1, '')"}
        ],
    )

[mwouts]

Indeed, you're right! The linked example uses JS code rather than a JS function.

I did some experiment with a new JavascriptCode class in the columns_render branch, and I have had some success with this:

pip install git+https://github.com/mwouts/itables.git@columns_render

and then

import pandas as pd
from itables import show
from itables import JavascriptCode

import math
show(pd.Series([i * math.pi * 1e4 for i in range(1, 6)]), columnDefs=[
        {
            "targets": "_all",
            "render": JavascriptCode("$.fn.dataTable.render.number(',', '.', 3, '$')"),
        }
    ])

This worked in Jupyter Lab, but it did not work in Jupyter Book (the JS errors are $.fn.dataTable is undefined and $.getQueryParameters is not a function).

This is very experimental and I have the feeling that passing a function rather than JS code might work better. I see here that render can as well be a function. @AllanJard, is it easy to write a function that applies the same formatting as dataTable.render.number?

[AllanJard]

The code for it is here if you wanted to basically the same thing.

[mck-star-yar]

@mwouts, the code works perfectly. Thanks a lot for such quick help! What is the ETA for this feature to get into the main branch?

[mwouts]

Thank you for your feedback! I just got the precious help of @fwouts and we were able to make this work for both the connected and the offline mode, so we're good to go...