Seaborn: Add datetime support to regression plots

[BartekPog]

Idea

Regression plots in time series are useful to create basic overviews of the data changes and levels. The example use case has been presented in this repository.

Plotting regression plots like relplot or lmplot with dateTime as the x axis is currently not supported. It results in an error.

TypeError: unsupported operand type(s) for *: 'Timestamp' and 'float'

Full error log

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
~/code/data-analysis/yt-history/exploration.py in 
----> 35 sns.lmplot(data=grouped, x="date", y="title", hue="language", order=4,scatter=False)

~/code/data-analysis/yt-history/.env/lib/python3.8/site-packages/seaborn/_decorators.py in inner_f(*args, **kwargs)
     44             )
     45         kwargs.update({k: arg for k, arg in zip(sig.parameters, args)})
---> 46         return f(**kwargs)
     47     return inner_f
     48 

~/code/data-analysis/yt-history/.env/lib/python3.8/site-packages/seaborn/regression.py in lmplot(x, y, data, hue, col, row, palette, col_wrap, height, aspect, markers, sharex, sharey, hue_order, col_order, row_order, legend, legend_out, x_estimator, x_bins, x_ci, scatter, fit_reg, ci, n_boot, units, seed, order, logistic, lowess, robust, logx, x_partial, y_partial, truncate, x_jitter, y_jitter, scatter_kws, line_kws, size)
    623         scatter_kws=scatter_kws, line_kws=line_kws,
    624     )
--> 625     facets.map_dataframe(regplot, x=x, y=y, **regplot_kws)
    626 
    627     # TODO this will need to change when we relax string requirement

~/code/data-analysis/yt-history/.env/lib/python3.8/site-packages/seaborn/axisgrid.py in map_dataframe(self, func, *args, **kwargs)
    748 
    749             # Draw the plot
--> 750             self._facet_plot(func, ax, args, kwargs)
    751 
    752         # Finalize the annotations and layout

~/code/data-analysis/yt-history/.env/lib/python3.8/site-packages/seaborn/axisgrid.py in _facet_plot(self, func, ax, plot_args, plot_kwargs)
    772                 plot_kwargs[key] = val
    773             plot_args = []
--> 774         func(*plot_args, **plot_kwargs)
    775 
    776         # Sort out the supporting information

~/code/data-analysis/yt-history/.env/lib/python3.8/site-packages/seaborn/_decorators.py in inner_f(*args, **kwargs)
     44             )
     45         kwargs.update({k: arg for k, arg in zip(sig.parameters, args)})
---> 46         return f(**kwargs)
     47     return inner_f
     48 

~/code/data-analysis/yt-history/.env/lib/python3.8/site-packages/seaborn/regression.py in regplot(x, y, data, x_estimator, x_bins, x_ci, scatter, fit_reg, ci, n_boot, units, seed, order, logistic, lowess, robust, logx, x_partial, y_partial, truncate, dropna, x_jitter, y_jitter, label, color, marker, scatter_kws, line_kws, ax)
    833     scatter_kws["marker"] = marker
    834     line_kws = {} if line_kws is None else copy.copy(line_kws)
--> 835     plotter.plot(ax, scatter_kws, line_kws)
    836     return ax
    837 

~/code/data-analysis/yt-history/.env/lib/python3.8/site-packages/seaborn/regression.py in plot(self, ax, scatter_kws, line_kws)
    368 
    369         if self.fit_reg:
--> 370             self.lineplot(ax, line_kws)
    371 
    372         # Label the axes

~/code/data-analysis/yt-history/.env/lib/python3.8/site-packages/seaborn/regression.py in lineplot(self, ax, kws)
    411         """Draw the model."""
    412         # Fit the regression model
--> 413         grid, yhat, err_bands = self.fit_regression(ax)
    414         edges = grid[0], grid[-1]
    415 

~/code/data-analysis/yt-history/.env/lib/python3.8/site-packages/seaborn/regression.py in fit_regression(self, ax, x_range, grid)
    199                 else:
    200                     x_min, x_max = ax.get_xlim()
--> 201             grid = np.linspace(x_min, x_max, 100)
    202         ci = self.ci
    203 

<__array_function__ internals> in linspace(*args, **kwargs)

~/code/data-analysis/yt-history/.env/lib/python3.8/site-packages/numpy/core/function_base.py in linspace(start, stop, num, endpoint, retstep, dtype, axis)
    118     # Convert float/complex array scalars to float, gh-3504
    119     # and make sure one can use variables that have an __array_interface__, gh-6634
--> 120     start = asanyarray(start) * 1.0
    121     stop  = asanyarray(stop)  * 1.0
    122 

TypeError: unsupported operand type(s) for *: 'Timestamp' and 'float'

Solution

If the x axis is an instance of Timestamp convert it to float before fitting the regression model.

[mwaskom]

This is on the roadmap — half of the work is done with the new core features that permit histogram/KDE over datetimes as seen in 0.11.0. But now the regression plotter needs to be refactored to use that machinery.

[mwaskom]

In the meantime, it's relatively easy to achieve externally, something like

x = np.random.choice(np.arange("2010-01-01", "2020-01-01", dtype="datetime64[D]"), 200)
y = np.random.normal(size=len(x))

f, ax = plt.subplots()
ax.xaxis.update_units(x)
sns.regplot(x=ax.xaxis.convert_units(x), y=y)