Reputation
16,610
Next tag badge:
90/100 score
23/20 answers
Badges
7 46 95
Newest
 Good Answer
Impact
~1.6m people reached

2d
comment Is it necessary to detrend and decycle time-series data when using machine learning methods?
As I wrote: 1/ ML is bad at extrapolation, 2/ time uniquely identifies each object, which allows severe overfitting and inhibits pursuit for actual interactions.
Jan
31
awarded  Good Answer
Jan
30
revised R LASSO always include some coefficient and question about data partition
Code as code.
Jan
25
revised What is behind Google Prediction API?
Updated URL, previous was broken.
Jan
19
comment Free data set for very high dimensional classification
It is a sparse array, integer N means attribute N value is 1.
Jan
14
revised Linear regression on choccake dataset
added 13 characters in body
Jan
14
revised Linear regression on choccake dataset
added 6 characters in body; edited title
Jan
14
answered Linear regression on choccake dataset
Jan
9
answered Accuracy of training sample in Random Forest model in R
Jan
5
answered Testing multiple Classification models
Dec
15
comment How to tune the “depth” and “min_samples_leaf” of Random Forest with correlated data?
Possibly; yet the only way to be sure is to just make few tests.
Dec
15
comment How does feature selection work in Random Forest?
The link is dead...
Dec
15
answered How to tune the “depth” and “min_samples_leaf” of Random Forest with correlated data?
Dec
15
reviewed Approve How to tune the “depth” and “min_samples_leaf” of Random Forest with correlated data?
Dec
8
answered Classification problem with statistically insignificant variables
Dec
7
comment Test that score isn't predicted by chance
In a most basic approach you simply need some kind of zero hypothesis, i.e., a way to make a random (yet representative wrt your real networks) graph in which adding any more edge would be an error. Then just apply your method to many such graphs and gather the distribution of score expected by chance.
Dec
2
awarded  Nice Answer
Nov
30
revised Generalising correlation-based feature selection
added 4 characters in body; edited title
Nov
30
revised Why are correlated observations not an issue in Cox regression with time-dependent covariates?
Formatting, LaTeX.
Nov
29
comment When will L1 regularization work better than L2 and vice versa?
Like those enumerated here.