modeloriented / live Goto Github PK

Local Interpretable (Model-agnostic) Visual Explanations - model visualization for regression problems and tabular data based on LIME method. Available on CRAN

Home Page: https://modeloriented.github.io/live/

License: Other

R 100.00%

model-visualization machine-learning interpretability lime visual-explanations xai iml

live's People

Contributors

Stargazers

Watchers

Forkers

radovankavicky gapdata guhjy mdwahiduzzamanemon summerflowers

live's Issues

Make selection work for different link functions

some error in sample_locally

In the wine_quality() vigniette call

similar <- sample_locally(data = winequality_red,
                            explained_instance = winequality_red[5, ], 
                            explained_var = "quality", 
                            size = 100,
                            standardise = TRUE)

results

Error in mutate_impl(.data, dots) : 
  Evaluation error: object 'x' not found.

Naprawić ctree

'valid.viewport(x, y, width, height, just, gp, clip, xscale, yscale, ':
invalid 'yscale' in viewport

Error in installation

devtools::install_github("MI2DataLab/live")
Downloading GitHub repo MI2DataLab/live@master
from URL https://api.github.com/repos/MI2DataLab/live/zipball/master
Installing live
"C:/PROGRA~~1/R/R-34~~1.2/bin/x64/R" --no-site-file --no-environ
--no-save --no-restore --quiet CMD INSTALL
"C:/Users/Jinn-Yuh/AppData/Local/Temp/RtmpqCNN78/devtools1a6c27391bb/MI2DataLab-live-8a454c6"
--library="D:/Dropbox/Stat/R/Library" --install-tests

installing source package 'live' ...
Error : Invalid DESCRIPTION file

Malformed maintainer field.

See section 'The DESCRIPTION file' in the 'Writing R Extensions'
manual.

ERROR: installing package DESCRIPTION failed for package 'live'

Update pkgdown site

with the instruction

pkgdown::build_site()

Funkcja, która przyjmuje gotowy model

i na jego podstawie robi predykcję dla nowego zbioru danych.
Problem: model z mlr/nie z mlr.

Bardziej intuicyjne, krotsze nazwy

wide data - big problem

In the TCGA use case we have like 20000 predictors, this causes 2 types of problems:

ranger nor randomForest are not working for this number of features, so I am calling them on subset of 10k features
for white box classifiers we need more samples in the surroundings than dimensions. So the default 50 is far not enough (otherwise we will fall in the p >> n problem for the white box).
And 20k is too time-consuming.

ranger is not working with live

In the new use-case (https://github.com/MI2DataLab/live/blob/master/vignettes/TCGA.Rmd)
I was using ranger package to build quickly a forest.
But something is wrong with the predict interface. I am getting following error

> similar <- simulateSimilar(data = nData,
+                            newData = nData[1,], 
+                            blackBox = trees, 
+                            explainedVar = "survival_status", 
+                            size = 50)
Error in "gwaa.data" %in% class(data) : 
  argument "data" is missing, with no default

improve unit tests

factors in forestplot

value of factor at proper level needs to be added
levels can be coded better

Błąd w poleceniu 'sample_locally(data = winequality_red, explained_instance = winequality_red[5, ': Assertion on 'data' failed: Must be of type 'data.frame', not 'closure'

sample_locally(winequality_red,
winequality_red[5, ],
black_box = "regr.lm",
explained_var = "quality",
size = 100,
standardise = TRUE)

Nie działa

data <- winequality_red
sample_locally(data,
data[5, ],
black_box = "regr.lm",
explained_var = "quality",
size = 100,
standardise = TRUE)

Działa
Wcześniej nie było tego problemu

library(live)
dim(HR_data)
HR_data$left <- as.numeric(HR_data$left)
trees <- randomForest(left~., data = HR_data, ntree=1000)

similar <- sample_locally(data = HR_data,
                          explained_instance = HR_data[1,],
                          explained_var = "left", 
                          size = 2000)

similar1 <- add_predictions(HR_Data, similar, black_box_model = trees)

trained <- fit_explanation(live_object = similar1, 
                           white_box = "regr.lm", selection = FALSE)
Error in fit_explanation(live_object = similar1, white_box = "regr.lm",  : 
  Assertion on 'choices' failed. Must be of length >= 1, but has length 0.

Dokumentacja

przyklady