📍 Interactive Studio for Explanatory Model Analysis
modelStudio()
: open_plots = c("fi")
, which is a vector listing plots to be initially opened (and on which positions) (#112)
DALEX::loss_default()
since DALEX >=2.5.0
SV
to the same as in BD
modelStudio()
: max_features_fi = max_features
, which allows displaying a distinctive number of features in FI
plot (other than in BD
and SV
)ms_options()
: **_axis_title
, which allow changing plot-specific axis title (default varies)R >=3.6
, DALEX >=2.2.1
, ingredients >=2.2.0
and iBreakDown >=2.0.1
modelStudioOptions()
is removed from this version of the package; after being deprecated for over a year since v1.1.0. Use the recommended ms_options()
instead.modelStudio()
: N_sv = 3*N
, which by default decreases the number of observations used for the calculation of Shapley Values
(rows in data
)margin_left = NULL
by default and it is adjusted based on the length of variable namesFI
instead of BD
by defaultverbose
parameter to modelStudio()
as an alias to show_info
(#101)
ms_merge_observations()
function that merges local explanation of observations from multiple modelStudio
objects (#102)
modelStudio()
when data had only one variable (#99)
new_observation = NULL
then choose new_observation_n = 3
observations, evenly spread by the order of y_hat
. This shall always include the observations, which ids are which.min(y_hat)
and which.max(y_hat)
. Additionally, improve the observation dropdown text in dashboard. (#94)
DALEX v2.0.1
ms_options
: ms_subtitle
, ms_margin_top
and ms_margin_bottom
modelStudio()
: N_fi = 10*N
and B_fi = B
license
parameter to modelStudio()
which allows to specify the connection for readLines()
(e.g. 'LICENSE'
) which will add file contents into the HTML output as a commentDALEX v2.0
, ingredients v2.0
and iBreakDown v1.3.1
telemetry
to FALSE
.modelStudio
in Shiny (#77) using new widget_id
argumentNA
in data
(#71)
variable_splits_type='uniform'
and CP plots are now calculated with variable_splits_with_obs=TRUE
(#74)
loss_function
in FI is now different for each explainer$model_info$type
(#73)
...
would cause an errormax_vars
alias for the max_features
parametery
has only two unique values (classification) (#76)
stringsAsFactors=TRUE
where data.frame
is usedmodelStudioOptions()
to ms_options()
ms_update_options()
function that updates the options of a modelStudio
objectms_update_observations()
function that updates the observations of a modelStudio
objectB
default value from 15
to 10
and N
default value from 400
to 300
feature_importance
is now calculated on 10*N
sampled rows from the dataranger
instead of randomForest
everywhereauto_unbox = TRUE
to jsonlite::toJSON
and changed the .js
code to comply"modelStudio"
to the modelStudio()
outputFD
plot on matrix-like datamodelStudio
position in HTMLmodelStudio
now renders properly from .Rmd
to .html
R v4.0
changesdata
with xgboost
working examplemax(min, q1 - 1.5*iqr)
and min(max, q3 + 1.5*iqr)
show_info
with progress
package