14,553 reputation
63786
bio website mbq.me
location Warsaw, Poland
age 28
visits member for 4 years, 6 months
seen 20 mins ago


19m
revised What is your favorite “data analysis” cartoon?
deleted 45 characters in body
1d
comment Is it better to use MAE or MSE for perfomance measure?
The point is that you make such plots for RF and SVM, go to your forest people and discuss which distribution of error suits them better. This way you will do a rational choice between methods, not some random-people-on-the-internet-told-me-that-one-number-is-more-magic-than-other thing.
2d
comment Is it better to use MAE or MSE for perfomance measure?
The code for the example plot is just plot(abs(predicted-true)~cut2(true,g=10)); only the Hmisc package is needed to be installed and loaded first.
2d
revised Is it better to use MAE or MSE for perfomance measure?
added 114 characters in body
2d
answered Is it better to use MAE or MSE for perfomance measure?
Jan
15
comment K-fold validation, how to use MSE and STD for model selection
You can do a Wilcoxon test or t-test (after appropriate normal distribution check) using per-fold MSE sets as compared samples; it is likely that it will tell you that models are statistically indistinguishable, thus that model selection in this context is pointless.
Jan
15
awarded  Revival
Jan
15
revised Is accuracy = 1- test error rate
edited tags; edited title
Jan
15
answered Is accuracy = 1- test error rate
Jan
14
answered K-fold validation, how to use MSE and STD for model selection
Jan
14
revised Correlation between ordinal and average values
deleted 8 characters in body; edited title
Jan
12
answered Boruta score goes to minus infinity
Dec
2
awarded  Good Question
Nov
29
comment fragmentation problem in decision tree
Note that decision tree and decision tree ensembles are two different worlds -- fragmentation (basically noise in the near-leaf branches) in ensemble should average out. Single decision tree is mostly about making interpretable and meaningful model, not necessarily perfectly accurate.
Nov
21
awarded  Nice Answer
Nov
16
revised Selecting kernel or binary similarity measures
edited body; edited title
Nov
16
revised Identifiability and estimability
edited body; edited title
Nov
16
revised Classification algorithms that return confidence?
edited title
Nov
16
comment Test accuracy higher than training. How to interpret?
From definition, ML methods are about performance on unseen data, thus they give no guarantees about this result (the model is not expected to re-implement the mechanism underlying the data like in statistical modelling). In practice many methods give overly accurate predictions, thus it is only deceiving.
Nov
14
awarded  Good Answer