trouble with downloading rtemis

[amin0511ss]

Hi there I am having trouble with downloading the package. I did try to troubleshoot using available sites and comments/solutions discussed by other users but still am unable to download. the message is error shown below:

• installing source package 'rtemis' ... using staged installation R Error in parse(outFile) : C:/Users/...../AppData/Local/Temp/RtmpG2toXY/R.INSTALL9e6022be7db0/rtemis/R/dplot3.bar.R:209:16: unexpected '>' 208: for (i in seq(ncol(dat))) { 209: plt |> ^ ERROR: unable to collate and parse R files for package 'rtemis'
• removing 'C:/Users/....../Documents/R/R-4.0.3/library/rtemis'

thank you kindly

[egenn]

Hi, @amin0511ss The error suggests you are using R version < 4.1. Try updating to the latest version.

[amin0511ss]

Thank you Stathis for your quick response. I will try. I hope i can make good use of your great package. By the way, does your package have the capability to run "super" models (ensembles) like SuperLearner package? regards, amin

On Thu, Nov 4, 2021 at 4:13 PM Stathis Gennatas @.***> wrote:

Hi, @amin0511ss https://github.com/amin0511ss The error suggests you are using R version < 4.1. Try updating to the latest version.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/egenn/rtemis/issues/34#issuecomment-961382163, or unsubscribe https://github.com/notifications/unsubscribe-auth/AM5IEESYUQTPX2TTI23UECLUKLZP7ANCNFSM5HMIZ4YA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

[egenn]

I added a chapter in the documentation on meta models here: https://rtemis.lambdamd.org/metamodels.html For biomedical problems where we care about interpretability and not just performance, we usually train a suite of models and compare performance metrics and variable importance, instead of using stacking/blending/superlearners/meta-models.

[amin0511ss]

Thank you. Makes sense. I am yet to see a stacked up model or super model meaningfully being better than individual models. I will read the chapter thank you. I was playing around with your resample function and even when I do kfold I still see that there are overlaps between folds. Not sure what I am doing wrong. I want to do kfold with no overlap and then bootstrap within each fold to bring the numbers back to original count.

On Sat, Nov 6, 2021 at 12:18 AM Stathis Gennatas @.***> wrote:

I added a chapter in the documentation on meta models here: https://rtemis.lambdamd.org/metamodels.html For biomedical problems where we care about interpretability and not just performance, we usually train a suite of models and compare performance metrics and variable importance, instead of using stacking/blending/superlearners/meta-models.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/egenn/rtemis/issues/34#issuecomment-962410831, or unsubscribe https://github.com/notifications/unsubscribe-auth/AM5IEEVIYUO4MW46PMAJWZLUKTQETANCNFSM5HMIZ4YA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

[amin0511ss]

Sorry for writing again. In my work dealing with imbalance data is always a challenge based on my experience downsampling of the majority class seems to work best even when compared with smote. I used the capability in randomforest package or caret to adjust for sampling. Cost sensitive learning for some reason is subpar when compared with under or over sampling. This is true for both bagging and boosting. Do you have a function in rtemis to specify how to sample for majority and minority class? Regards

On Thu, Nov 4, 2021 at 1:13 PM Stathis Gennatas @.***> wrote:

Hi, @amin0511ss https://github.com/amin0511ss The error suggests you are using R version < 4.1. Try updating to the latest version.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/egenn/rtemis/issues/34#issuecomment-961382163, or unsubscribe https://github.com/notifications/unsubscribe-auth/AM5IEESYUQTPX2TTI23UECLUKLZP7ANCNFSM5HMIZ4YA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

[egenn]

Thank you. Makes sense. I am yet to see a stacked up model or super model meaningfully being better than individual models. I will read the chapter thank you. I was playing around with your resample function and even when I do kfold I still see that there are overlaps between folds. Not sure what I am doing wrong. I want to do kfold with no overlap and then bootstrap within each fold to bring the numbers back to original count. … On Sat, Nov 6, 2021 at 12:18 AM Stathis Gennatas @.***> wrote: I added a chapter in the documentation on meta models here: https://rtemis.lambdamd.org/metamodels.html For biomedical problems where we care about interpretability and not just performance, we usually train a suite of models and compare performance metrics and variable importance, instead of using stacking/blending/superlearners/meta-models. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#34 (comment)>, or unsubscribe https://github.com/notifications/unsubscribe-auth/AM5IEEVIYUO4MW46PMAJWZLUKTQETANCNFSM5HMIZ4YA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

Can you provide an example?

[egenn]

Sorry for writing again. In my work dealing with imbalance data is always a challenge based on my experience downsampling of the majority class seems to work best even when compared with smote. I used the capability in randomforest package or caret to adjust for sampling. Cost sensitive learning for some reason is subpar when compared with under or over sampling. This is true for both bagging and boosting. Do you have a function in rtemis to specify how to sample for majority and minority class? Regards … On Thu, Nov 4, 2021 at 1:13 PM Stathis Gennatas @.***> wrote: Hi, @amin0511ss https://github.com/amin0511ss The error suggests you are using R version < 4.1. Try updating to the latest version. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#34 (comment)>, or unsubscribe https://github.com/notifications/unsubscribe-auth/AM5IEESYUQTPX2TTI23UECLUKLZP7ANCNFSM5HMIZ4YA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

You can choose between inverse probability weighting (default), upsampling and downsampling. Docs: https://rtemis.lambdamd.org/imbalanced.html

[egenn]

Thank you. Makes sense. I am yet to see a stacked up model or super model meaningfully being better than individual models. I will read the chapter thank you. I was playing around with your resample function and even when I do kfold I still see that there are overlaps between folds. Not sure what I am doing wrong. I want to do kfold with no overlap and then bootstrap within each fold to bring the numbers back to original count. … On Sat, Nov 6, 2021 at 12:18 AM Stathis Gennatas @.***> wrote: I added a chapter in the documentation on meta models here: https://rtemis.lambdamd.org/metamodels.html For biomedical problems where we care about interpretability and not just performance, we usually train a suite of models and compare performance metrics and variable importance, instead of using stacking/blending/superlearners/meta-models. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#34 (comment)>, or unsubscribe https://github.com/notifications/unsubscribe-auth/AM5IEEVIYUO4MW46PMAJWZLUKTQETANCNFSM5HMIZ4YA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

Can you provide an example?

# 50 cases
x <- 1:50
# 10-fold
kfold <- resample(x, n.resamples = 10, resampler = "kfold")
# plot: teal is training, black is testing
plot(kfold)
# matrix with one line per resample; '1' if case is testing, '0' otherwise
mat <- matrix(1, 10, 50)
for (i in 1:10) mat[i, kfold[[i]]] <- 0
# how many times each case is used for testing across 10 folds
colSums(mat)

[amin0511ss]

thank you again. all very helpful. It makes sense: the test observations across folds have no overlap. I will try with three approaches (up, down, and inverse probability weight) and compare the result. I assume that the tuning of the hyperparameters is done in the background and when i print the model it would say which setup is used. for example, for RF, mtry and nodesize (and ntree) is available for the optimized model? regards,

On Sat, Nov 6, 2021 at 2:46 PM Stathis Gennatas @.***> wrote:

Thank you. Makes sense. I am yet to see a stacked up model or super model meaningfully being better than individual models. I will read the chapter thank you. I was playing around with your resample function and even when I do kfold I still see that there are overlaps between folds. Not sure what I am doing wrong. I want to do kfold with no overlap and then bootstrap within each fold to bring the numbers back to original count. … <#m7289099102689159139> On Sat, Nov 6, 2021 at 12:18 AM Stathis Gennatas @.***> wrote: I added a chapter in the documentation on meta models here: https://rtemis.lambdamd.org/metamodels.html For biomedical problems where we care about interpretability and not just performance, we usually train a suite of models and compare performance metrics and variable importance, instead of using stacking/blending/superlearners/meta-models. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#34 (comment) https://github.com/egenn/rtemis/issues/34#issuecomment-962410831>, or unsubscribe https://github.com/notifications/unsubscribe-auth/AM5IEEVIYUO4MW46PMAJWZLUKTQETANCNFSM5HMIZ4YA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub .

Can you provide an example?

50 cases

x <- 1:50

10-fold

kfold <- resample(x, n.resamples = 10, resampler = "kfold")

plot: teal is training, black is testing

plot(kfold)

matrix with one line per resample; '1' if case is testing, '0' otherwise

mat <- matrix(1, 10, 50) for (i in 1:10) mat[i, kfold[[i]]] <- 0

how many times each case is used for testing across 10 folds

colSums(mat)

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/egenn/rtemis/issues/34#issuecomment-962493808, or unsubscribe https://github.com/notifications/unsubscribe-auth/AM5IEETOIMEE43PHICCK6TTUKWAY5ANCNFSM5HMIZ4YA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

[amin0511ss]

Hi Stathis, I tried RF and RANGER for my data. For some reason, when i try to plot variable importance for RF i get an error message (i would not have any problem plotting variable importance for RANGER). also i did not see any mention of what the importance plot is based on (permutation, impurity decrease, SHAP, etc.) also i was wondering what the values of the x-axis for the importance variable plot is (for example, what does 80 mean? is it normalized value of something?) I also tried to find out how i could print the hyperparameter setup of the tuned model but i could not find it on your site. Could you kindly advise me? I also can't get the curve on ROC plot. i get the warning message (shown at the end) that font family not found in windows font base.

[image: image.png]

[image: image.png] [image: image.png]

On Sat, Nov 6, 2021 at 2:46 PM Stathis Gennatas @.***> wrote:

Thank you. Makes sense. I am yet to see a stacked up model or super model meaningfully being better than individual models. I will read the chapter thank you. I was playing around with your resample function and even when I do kfold I still see that there are overlaps between folds. Not sure what I am doing wrong. I want to do kfold with no overlap and then bootstrap within each fold to bring the numbers back to original count. … <#m7289099102689159139> On Sat, Nov 6, 2021 at 12:18 AM Stathis Gennatas @.***> wrote: I added a chapter in the documentation on meta models here: https://rtemis.lambdamd.org/metamodels.html For biomedical problems where we care about interpretability and not just performance, we usually train a suite of models and compare performance metrics and variable importance, instead of using stacking/blending/superlearners/meta-models. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#34 (comment) https://github.com/egenn/rtemis/issues/34#issuecomment-962410831>, or unsubscribe https://github.com/notifications/unsubscribe-auth/AM5IEEVIYUO4MW46PMAJWZLUKTQETANCNFSM5HMIZ4YA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub .

Can you provide an example?

50 cases

x <- 1:50

10-fold

kfold <- resample(x, n.resamples = 10, resampler = "kfold")

plot: teal is training, black is testing

plot(kfold)

matrix with one line per resample; '1' if case is testing, '0' otherwise

mat <- matrix(1, 10, 50) for (i in 1:10) mat[i, kfold[[i]]] <- 0

how many times each case is used for testing across 10 folds

colSums(mat)

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/egenn/rtemis/issues/34#issuecomment-962493808, or unsubscribe https://github.com/notifications/unsubscribe-auth/AM5IEETOIMEE43PHICCK6TTUKWAY5ANCNFSM5HMIZ4YA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

[egenn]

Hi - None of your images are showing.

I pushed a few updates:

• s.RF includes variable importance
• plots use OS default fonts to avoid problems on Windows - previously set to Helvetica, so should have worked in most cases
• s.RANGER documentation for "importance" options: "none", "impurity", "impurity_corrected", or "permutation".

For variable importance measures see package documentation, e.g. ranger::ranger s.RANGER is preferred over s.RF because it's faster / parallelized

If mod is your trained ranger model, mod$extra$gridSearch contains all the grid search output if the model was tuned. (Plan is to move that to mod$gridSearch at some point soon probably)

[amin0511ss]

hi again, here is a screenshot of where parameter values should be but it shows "NULL" value

[image: image.png]

On Sat, Nov 6, 2021 at 2:46 PM Stathis Gennatas @.***> wrote:

Thank you. Makes sense. I am yet to see a stacked up model or super model meaningfully being better than individual models. I will read the chapter thank you. I was playing around with your resample function and even when I do kfold I still see that there are overlaps between folds. Not sure what I am doing wrong. I want to do kfold with no overlap and then bootstrap within each fold to bring the numbers back to original count. … <#m7289099102689159139> On Sat, Nov 6, 2021 at 12:18 AM Stathis Gennatas @.***> wrote: I added a chapter in the documentation on meta models here: https://rtemis.lambdamd.org/metamodels.html For biomedical problems where we care about interpretability and not just performance, we usually train a suite of models and compare performance metrics and variable importance, instead of using stacking/blending/superlearners/meta-models. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#34 (comment) https://github.com/egenn/rtemis/issues/34#issuecomment-962410831>, or unsubscribe https://github.com/notifications/unsubscribe-auth/AM5IEEVIYUO4MW46PMAJWZLUKTQETANCNFSM5HMIZ4YA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub .

Can you provide an example?

50 cases

x <- 1:50

10-fold

kfold <- resample(x, n.resamples = 10, resampler = "kfold")

plot: teal is training, black is testing

plot(kfold)

matrix with one line per resample; '1' if case is testing, '0' otherwise

mat <- matrix(1, 10, 50) for (i in 1:10) mat[i, kfold[[i]]] <- 0

how many times each case is used for testing across 10 folds

colSums(mat)

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/egenn/rtemis/issues/34#issuecomment-962493808, or unsubscribe https://github.com/notifications/unsubscribe-auth/AM5IEETOIMEE43PHICCK6TTUKWAY5ANCNFSM5HMIZ4YA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

[amin0511ss]

thank you sir. i saw that i had to include grid points in the call. i assume all the names of the hyperparameters are identical to the original algorithms. for example, for randomforest it would be mtry, nodesize, etc. thank you again. your package is a game changer for me mod.rf.bal <- s.RF(dat.train, dat.test, ipw = FALSE, downsample = TRUE,mtry=seq(3,7,1),nodesize = c(1,2,5))

On Sat, Nov 6, 2021 at 7:54 PM Stathis Gennatas @.***> wrote:

Hi - None of your images are showing.

I pushed a few updates:

• s.RF includes variable importance
• plots use OS default fonts to avoid problems on Windows - previously set to Helvetica, so should have worked in most cases
• s.RANGER documentation for "importance" options: "none", "impurity", "impurity_corrected", or "permutation".

For variable importance measures see package documentation, e.g. ranger::ranger s.RANGER is preferred over s.RF because it's faster / parallelized

If mod is your trained ranger model, mod$extra$gridSearch contains all the grid search output if the model was tuned. (Plan is to move that to mod$gridSearch at some point soon probably)

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/egenn/rtemis/issues/34#issuecomment-962525654, or unsubscribe https://github.com/notifications/unsubscribe-auth/AM5IEEQL4S2CAV55VCNAEDLUKXE3BANCNFSM5HMIZ4YA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

[egenn]

I think most if not all arguments are the same as original package, but check function usage and arguments to verify name and whether they are included in the automatic grid search as indicated by "[gS]" in the documentation

[amin0511ss]

hi, i re-installed your package but still am unable to plot varimp for s.RF model and same with plotROC (still gives the same warning)

On Sat, Nov 6, 2021 at 7:54 PM Stathis Gennatas @.***> wrote:

Hi - None of your images are showing.

I pushed a few updates:

• s.RF includes variable importance
• plots use OS default fonts to avoid problems on Windows - previously set to Helvetica, so should have worked in most cases
• s.RANGER documentation for "importance" options: "none", "impurity", "impurity_corrected", or "permutation".

For variable importance measures see package documentation, e.g. ranger::ranger s.RANGER is preferred over s.RF because it's faster / parallelized

If mod is your trained ranger model, mod$extra$gridSearch contains all the grid search output if the model was tuned. (Plan is to move that to mod$gridSearch at some point soon probably)

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/egenn/rtemis/issues/34#issuecomment-962525654, or unsubscribe https://github.com/notifications/unsubscribe-auth/AM5IEEQL4S2CAV55VCNAEDLUKXE3BANCNFSM5HMIZ4YA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

[amin0511ss]

Thank you for the clarification

On Sat, Nov 6, 2021 at 5:18 PM Stathis Gennatas @.***> wrote:

I think most if not all arguments are the same as original package, but check function usage and arguments to verify name and whether they are included in the automatic grid search as indicated by "[gS]" in the documentation

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/egenn/rtemis/issues/34#issuecomment-962528618, or unsubscribe https://github.com/notifications/unsubscribe-auth/AM5IEEXYWNXUBWJZ3UKNGBLUKXHVJANCNFSM5HMIZ4YA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

[egenn]

hi, i re-installed your package but still am unable to plot varimp for s.RF model and same with plotROC (still gives the same warning)

Your attachments aren't showing, I don't know what the message was, but make sure you've restarted your R session.

[amin0511ss]

thank you will do..thank you so much for being so responsive.

On Sat, Nov 6, 2021 at 8:27 PM Stathis Gennatas @.***> wrote:

hi, i re-installed your package but still am unable to plot varimp for s.RF model and same with plotROC (still gives the same warning)

Your attachments aren't showing, I don't know what the message was, but make sure you've restarted your R session.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/egenn/rtemis/issues/34#issuecomment-962529538, or unsubscribe https://github.com/notifications/unsubscribe-auth/AM5IEEXEYBPFXPKH5GK26BDUKXIY3ANCNFSM5HMIZ4YA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

[amin0511ss]

after the restart i still have the same problem

mod.rf.bal$plotVarImp() Warning message: In mod.rf.bal$plotVarImp() : Variable importance is not available for this model

mod.ranger.bal$plotROC() There were 16 warnings (use warnings() to see them)

warnings() Warning messages: 1: In axis(side = theme$x.axis.side, line = theme$x.axis.line, ... : font family not found in Windows font database

On Sat, Nov 6, 2021 at 8:27 PM Stathis Gennatas @.***> wrote:

hi, i re-installed your package but still am unable to plot varimp for s.RF model and same with plotROC (still gives the same warning)

Your attachments aren't showing, I don't know what the message was, but make sure you've restarted your R session.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/egenn/rtemis/issues/34#issuecomment-962529538, or unsubscribe https://github.com/notifications/unsubscribe-auth/AM5IEEXEYBPFXPKH5GK26BDUKXIY3ANCNFSM5HMIZ4YA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

[egenn]

It doesn't seem like you have the latest update

[amin0511ss]

Hi Stathis, I hope you are doing well.

1. Does "elevate" only work with "ranger"?
2. Is the "variable importance" plot only available within "elevate" (after applying elevate)?
3. What is the variable importance based on (permutation, impurity decrease, model-agnostic measures, etc.)? thank you so much sir amin

On Sat, Nov 6, 2021 at 9:45 PM Stathis Gennatas @.***> wrote:

It doesn't seem like you have the latest update

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/egenn/rtemis/issues/34#issuecomment-962537383, or unsubscribe https://github.com/notifications/unsubscribe-auth/AM5IEERVLYWEAJSGLMASDFLUKXR27ANCNFSM5HMIZ4YA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

[egenn]

Take a look at the package documentation and the online docs and vignettes. from ?elevate:

mod        Character: Learner to use. Options: modSelect

Re: variable importance: have you looked at the output? It's the same, $plotVarImp(). Look at individual learner functions for what variable importance means each time - if it's not there, you can look at the code.

Also note that the "issues" section is for code bugs and problems, not usage.

[amin0511ss]

thank you. When i click on "modSelect" hyperlink, it takes me to an error page that says "Oops, that page does not exist". I will look at the code. regards

On Tue, Nov 16, 2021 at 11:43 PM Stathis Gennatas @.***> wrote:

Take a look at the package documentation and the online docs and vignettes. from ?elevate:

mod Character: Learner to use. Options: modSelect

Re: variable importance: have you looked at the output? It's the same, $plotVarImp(). Look at individual learner functions for what variable importance means each time - if it's not there, you can look at the code.

Also note that the "issues" section is for code bugs and problems, not usage.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/egenn/rtemis/issues/34#issuecomment-971176746, or unsubscribe https://github.com/notifications/unsubscribe-auth/AM5IEERCCBNCV67ACCEZPEDUMMXGXANCNFSM5HMIZ4YA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

[egenn]

thank you. When i click on "modSelect" hyperlink, it takes me to an error page that says "Oops, that page does not exist".

That's not normal, modSelect is a documented function. Was the package installed correctly? Just run modSelect()