Plotly Express column input formats

[nicolaskruchten]

Right now for all px methods, data_frame is a required argument and the values of many kwargs are column names as strings.

We should be much more flexible in what we accept, for example the following should work:

px.scatter(x=[1,2,3], y=[1,2,3])

In which case we would internally build a data frame with column names "x" and "y".

In addition, the following should work:

px.scatter(df, x=df.index, y=df.col)

In which case we can read the column names (and optionally index name?) from the data frame.

Finally, we should allow for mixing and matching:

px.scatter(df, x=df.index, y=df.col, color=[1,2,3])

[nicolaskruchten]

A couple of additional thoughts... The following should work, and result in a data frame internally with column names "x" and "y" because the parent data frame was not provided to px:

px.scatter(x=df.x_col, y=df.y_col)

Finally, just to be explicit, we should accept the same formats wherever we accept lists of column names, like in dimensions, hover_data and custom_data.

[nicolaskruchten]

This was a PR which implemented part of these requirements earlier this summer, but was held up by the then-impending repo shuffle: plotly/plotly_express#87

[emmanuelle]

Also take a look at xarray

[nicolaskruchten]

Re xarray I seem to recall reading that 1-dimensional arrays could have an "onboard" name, unlike pandas.Series objects, whose names are only available in a containing pandas.DataFrame... if this name is easy to read, it might be nice to use it instead of "x" in the case of px.scatter(x=<xarray vector with onboard name>) :)

[emmanuelle]

We could also accept px.scatter(x={'time':[1,2,3]}, y={'revenue':[1,2,3]}). What do you think?

[emmanuelle]

It would be a way to give names to columns easily. We can also use the labels argument for this.

[nicolaskruchten]

Yes, I think that px.scatter(x=[1,2,3], y=[1,2,3], labels={'x': 'time', 'y': 'money'}) would be the way here.

[nicolaskruchten]

depending on how it's implemented, the above should "just work" actually :)