todo.txt

add permutation importance, partial dependencefe curves
add feature engineering for explainable models
refactor pca and lda plots together
make univariate pca nad lda plots maybe
think about train-test-split
benchmark plotting and easy classifier
add anyclassifier

if categorical encoded as float (even int) like chas variable, imputation fails with dabl_missing which is a string.

FIXME integers need to be considered continuous again in preprocessor? duplicate columns? hmmm
we need to think about this!

types are different for feature engineering and visualization.
try to convert floats to ints as well?

# NEW? where's the other one?! damn...

only plot pdp if few classes, otherwise restrict to user-specified classes

calibration plots

select pdp plot features by some measure of variable importance?

refactor finding "interesting features" in a model-agnostic way?

partial dependence plot can't take advantage of fast method for now :(

categorical plot for regression: maybe use histogram thing instead of bar plots by default?
problem currently: some categories can be really small!

move examples all into one folder.

categorical, categorical, continuous.
How do I plot it?!
Can I plot it lol? mosaic is basically "just" two categorical and already tricky...
categorical vs continuous is stacked bar charts
quadrants of bar charts? (small multiple plot kinda)

interactivity:
for each plot, pick one of the equivalent plot types
allow dropping a plot
continuous vs categorical vs ordinal for a feature
for regression plots, pick categorical for hue
logarithmic axis vs not
size and alpha?


Website: focus on vis stuff, show example of changing parameters for exploration
add docs to plot for kwargs that get handed through?