Creating a visual schematic diagram for data wrangling workflow in `{datawizard}`

[IndrajeetPatil]

IMHO, the current README is quite dull and long-winded, and doesn't provide much insight into how this package can be useful for the users.

What we need is for it to feature a visual schematics like the following ones in our other popular high-level packages:

Of course, paging our in-house visualization wizard @DominiqueMakowski! 🪄

Needless to say, this is low-priority and if you think so necessary, we can definitely wait for the package to even mature further.

[DominiqueMakowski]

What would it contain? I can start a draft with powerpoint

[IndrajeetPatil]

@DominiqueMakowski How about something like this? (cc @bwiernik, @strengejacke, @mattansb, @etiennebacher)

Of course, there is a lot of room for improvement here. Specifically,

• I am not sure how to visually depict messy versus clean/tidy data. The only thing I could come up with was a hamper full of dirty clothes versus clean, folded clothes. Maybe others have better ideas.
• This includes no functions about "Data Properties". Is it important to include them?
• The list of functions I've included in the two columns is incomplete. Not sure how comprehensive we want to be here.

[DominiqueMakowski]

I can give it a go next week (do ping me then if you remember :)

The list of functions I've included in the two columns is incomplete.

It's okay not to be comprehensive otherwise we will be obsolete as soon as we add a new function, better perhaps to create like a wordcloud or something like that

[IndrajeetPatil]

Yeah, I agree. That's why I had put the ... in those columns. I don't think we need to be comprehensive, but we should definitely include the most important ones (filter, select, join, etc.).

[bwiernik]

I think a separate viz of data cleaning versus data summary functions would be good

[IndrajeetPatil]

@DominiqueMakowski It will be nice to have something like this in the JOSS paper.

[DominiqueMakowski]

Will do within the next couple of days

[DominiqueMakowski]

Would be nice to generate a wordcloud of the functions tho

[DominiqueMakowski]

Wordlist for wordclouds (https://www.wordclouds.com/):

• Preparation:

data_filter() data_select() data_to_long() data_to_wide() data_rotate() data_rename() data_relocate() data_join()

• Transformation:

standardize() normalize() center() degroup() winsorize() data_cut() data_recode() data_shift()

[IndrajeetPatil]

I want to wait for #57 and #197 to be resolved before we can include the following functions in the wordcloud:

data_cut() data_recode() data_shift()

We should avoid including any functions names in a publication that we are not sure will survive for long.

[DominiqueMakowski]

you're right, I'll come up with a diagram prototype nonetheless and then we can fine-tune the wordcloud

[DominiqueMakowski]

We can focus on the dirty clothes metaphor but it lacks some text at the bottom? (feel free to directly edit the powerpoint on the diagram branch!)

[IndrajeetPatil]

Thanks, Dom! This looks like a great start.

I think one way this can be improved is by making it visually less busy and more minimal. Additionally, we need to mention only a few (key and most useful) functions and just have ... (which will cover all the other existing or future functions).

I don't like the star shape in the "Transformations" section.

Maybe this can be an ironing table with a shirt on it? As in, imperfections in prepared data are ironed out using statistical transformations before the data is ready to be fed into a statistical model.

Instead of "No dependencies", I'd write "Lightweight", since we do import{insight}.

[IndrajeetPatil]

I also want to hear what @etiennebacher, @strengejacke, @bwiernik, @mattansb think about the current status of the illustration and how it can be further improved.

[bwiernik]

I agree with Indra's comments and don't have much more to add there. I like the ironing metaphor (maybe the function names in a cloud of steam?). And agree that making the function names less busy/stand out more would be good

[mattansb]

Looks good. I would maybe change the color of the bg color of the washing machine to a lighter blue? And for transform use the non data_* variant names.

[etiennebacher]

Looks good to me too, but it's a bit hard to read most function names in steps 2 and 3. Maybe you can remove the very small ones to increase the size of the others?

[IndrajeetPatil]

Thank you all for great suggestions!

WDYT, @DominiqueMakowski? Will this be possible? Don't know how complicated it will be to design.

[strengejacke]

Before we finalize this, we should definitely decide on the new function names. Mostly, I'm not quite satisfied with change_code(). recode is a verb, and everyone would expect such a function would recode old into new values. in change_code(), the verb is change, and what can we expect if we change a code? What we actually change when recoding are values (or factor levels, but values is maybe more generic). What do you think about change_values()? or maybe recode_values(), or recode_variables().

[bwiernik]

The "code" being the mapping of quantities to values/labels. So the function is changing the coding scheme used.

Maybe change_coding()? change_values() would be my second choice

[IndrajeetPatil]

@DominiqueMakowski Let us know if these suggestions make sense.

[IndrajeetPatil]

bump

[strengejacke]

[IndrajeetPatil]

bump

[DominiqueMakowski]

Is that the correct list?

Preparation: data_filter() data_select() data_to_long() data_to_wide() data_rotate() data_rename() data_relocate() data_join()

Transformation: standardize() normalize() center() degroup() winsorize() categorize() change_code() slide()

[DominiqueMakowski]

thanks for the bumps 🙊

[IndrajeetPatil]

These need to change to their new names:

• data_cut() -> categorize()
• data_recode() -> recode_values() ~~change_code()~~
• data_shift() -> slide()

Btw, feel free to not include all of them. Whatever looks better with the chosen graphic design.

[bwiernik]

recode_values() not change_code()

[IndrajeetPatil]

bump