Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Improves data frame handling #1474

Open
wants to merge 3 commits into
base: main
Choose a base branch
from
Open

Improves data frame handling #1474

wants to merge 3 commits into from

Conversation

dfalbel
Copy link
Member

@dfalbel dfalbel commented Sep 6, 2023

Improves how the reticulate engine handles pandas data frames rendering.
This PR is an attempt to fix #783

In summary it adresses two problems:

  1. Quarto documents rendered with the Jupyter engine will by default use rich data.frame rendering, which is not the case if they use the reticulate/knitr engine.

    We prefer not having such difference between engines, so this PR will special case Quarto rendered documents and use .to_html() or ._repr_markdown() when appropriate when out-printing data.frames into documents in Quarto rendered documents.

  2. The reticulate engine doesn't honor RMarkdown's df_print option and always uses print(df) to display pandas data.frames.

    This PR adds support for such option and we now respect the df_print option when auto-printing pandas data.frames. When not using the 'default' value, we convert the pandas data frame into an R data frame and then delegate to knitr for printing.

Both changes are breaking changes, in the sense that they will now be the default. One should set reticulate.engine.render_df to FALSE in order to get back the old behavior.

@cderv If you have time, it would be awesome if you can take a look :)

Copy link
Contributor

@cderv cderv left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think this logic could be better than current for sure. Thanks !

Though several thoughts:

For Quarto, we could output Markdown or HTML. Quarto does this for Jupyter and I believe give priority HTML method; Example:

---
title: test
keep-md: true
---

```{python}
import pandas as pd
df = pd.DataFrame({"object": ["Sun", "Earth"], "radius": [696000, 6371]})
df
```

If you look at the intermediary .md output, the table is in its HTML representation.
However, Using Markdown representation could be the same as we parse HTML table
@cscheid do you think it should be like Quarto and Jupyter, or reticulate should output _repr_markdown_ by default for Pandas ? (regarding your comment at #783 (comment)
We can discuss what Quarto does exactly, but currently we select HTML - so I commented with that in mind.

For rmarkdown, I believe this is a good logic that will use any knit_print.data.frame method that another package could define. Is there any possibly loss passing by py_to_r() to rely on data.frame representation, instead of what Pandas HTML representation could do ?

R/knitr-engine.R Outdated
Comment on lines 693 to 695
if (knitr::opts_knit$get("rmarkdown.df_print") != "default" && renderDF) {
return(knitr::knit_print(py_to_r(value)))
}
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think we need to check really if using knit_print() at this level works really. I don't recall what knitr expect as output , but also it seems to me that there is a special mechanism in reticulate with .engine_context$pending_plots to store some content produce and then make it work with options in eng_python(). Next if (isHtml && py_has_method(value, "_repr_html_")) does something to output HTML raw data. I wonder if this is needed or not.

I don't know if you tested by at least, it is missing an argument here

Suggested change
if (knitr::opts_knit$get("rmarkdown.df_print") != "default" && renderDF) {
return(knitr::knit_print(py_to_r(value)))
}
if (knitr::opts_knit$get("rmarkdown.df_print") != "default" && renderDF) {
return(knitr::knit_print(py_to_r(value), options))
}

After this is fix it is working as expected it seems. (object is passed in captured and added to outputs$data() which is correctly passed to engine_output(). So looks good to me.

Copy link
Member Author

@dfalbel dfalbel Sep 12, 2023

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for the fix, I made the change in 8243db4

Simply returning the raw output seems to work correctly. knit_print output seems to have a knit_html class that is correctly handled by knitr later. DO you have ideas of what could the potential edge cases? I have tried things like printing multiple tables from the same chunk and it worked.

}

eng_python_generic_autoprint <- function(captured, value) {
if (py_has_method(value, "_repr_markdown_")) {
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

As commented in main comment, I wonder if we should do as Quarto and

  • Output HTML first if available
  • Output Markdown if available
  • Then captured (equivalent to text/plain)

Logic in Quarto for display is at displayMimeType() where we have a logic of ordering possibly output before checking the output values. In this logic, if there is HTML and Markdown representation, HTML is preferred currently

@dfalbel
Copy link
Member Author

dfalbel commented Sep 12, 2023

@cderv Thank you very much for the review, and sorry for taking so long to go back at this.

For rmarkdown, I believe this is a good logic that will use any knit_print.data.frame method that another package could define.

We could rely on knitr_print.pandas.core.frame.DataFrame and provide an implementation for it in reticulate. That would also affect the output of printing pandas data.frames from an R chunk, for instance the chunk below would use that method. Maybe this is a better option?

```{r}
reticulate::r_to_py(mtcars)
```

Is there any possibly loss passing by py_to_r() to rely on data.frame representation, instead of what Pandas HTML representation could do ?

py_to_r.data.frame() can be a lossy cast, but I'd guess users prefering the rich representation are fine with that, and are probably as the representation itself is also lossy if used with uncommon column types.

As commented in main comment, I wonder if we should do as Quarto and

Output HTML first if available
Output Markdown if available
Then captured (equivalent to text/plain)

I'm afraid I don't have enough information about pros and cons to decide. That can be a breaking change because the current engine behavior is to prefer the markdown representation. Maybe that's a good time to push this change? It would be nice to know though the benefits of prefering HTML over the markdown representation.

@joelostblom
Copy link
Contributor

Thanks for working on this! I'm currently writing a Quarto book that uses both dplyr and pandas to display dataframes as tables at several places in the book. Currently, I'm stuck at using hidden cells with ipython's HTML() function to display pandas dataframes . So far I'm having more success with using the jupyter kernel and rpy2, but there are still some issues there as well and I would like to explore knitr + reticulate more for writing bilingual textbooks. Are there any particular blockers for this PR or anything that can be done to move it further along?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

changing pandas dataframe display style in Rmarkdown
3 participants