User ihadanny

62 votes

4 answers

117k views

What is the proper usage of scale_pos_weight in xgboost for imbalanced datasets?

asked Oct 30, 2016 at 13:59

24 votes

4 answers

92k views

Why is P(A,B|C)/P(B|C) = P(A|B,C)?

asked Jan 26, 2017 at 22:13

19 votes

1 answer

8k views

Can you give a simple intuitive explanation of IRLS method to find the MLE of a GLM?

asked Sep 24, 2016 at 7:59

14 votes

1 answer

25k views

t.test returns an error "data are essentially constant"

asked Aug 23, 2014 at 5:33

9 votes

2 answers

1k views

Why is the variable importance metric suggested by Breiman specific only to random forests?

asked Sep 15, 2015 at 20:33

9 votes

2 answers

5k views

Can you give an intuition behind the FTRL update step?

asked Feb 4, 2016 at 9:58

9 votes

2 answers

8k views

Fisher's score function has mean zero - what does that even mean?

asked Aug 6, 2016 at 21:03

8 votes

1 answer

2k views

can you explicitly show me the first iteration of newton-raphson and fisher scoring?

asked Sep 23, 2016 at 20:20

7 votes

1 answer

126 views

need intuition: why is there no winning strategy to this game?

asked Mar 26, 2017 at 17:33

5 votes

2 answers

394 views

Why is it that when the score variance gets LARGER we get MORE confident about the MLE estimate?

asked Sep 9, 2016 at 12:11

4 votes

1 answer

142 views

In GLM do we try to model E(T(y)) or E(y)?

asked Aug 5, 2016 at 22:01

4 votes

1 answer

224 views

Should I avoid mixed ARMA models?

asked Sep 27, 2019 at 16:02

4 votes

1 answer

519 views

Feature importance for xgboost classification of a sample

asked Jan 4, 2017 at 10:11

4 votes

0 answers

139 views

when fitting a regression model to a time-series, can I use lagged values of the time-series itself?

asked Nov 13, 2019 at 12:46

3 votes

2 answers

280 views

using the test population as an eval_set when doing hyperparameter optimization

asked Jan 23, 2020 at 11:27

3 votes

0 answers

35 views

Why are these 2 ARIMA formulations equivalent?

asked Oct 31, 2019 at 20:39

3 votes

1 answer

124 views

Can we do hypothesis testing for OLS estimator without assuming normality for the errors?

asked Oct 22, 2016 at 18:19

3 votes

1 answer

3k views

Is it necessary to use warm_start when tracking oob_score in scikit RandomForestClassifier?

asked Mar 20, 2016 at 11:12

2 votes

0 answers

302 views

Interpreting random survival forest mortality

asked Jul 13, 2016 at 18:34

2 votes

0 answers

66 views

how come that the variance of an estimate is smaller than the variance of the true value?

asked Oct 23, 2015 at 18:05

2 votes

1 answer

585 views

time series forecasting - predicting the next 24 hours

asked Oct 3, 2019 at 5:42

2 votes

0 answers

592 views

LIME - local linear explanations for global black-box models work for LogisticRegression but not for RandomForest?

asked Apr 27, 2017 at 11:25

2 votes

1 answer

429 views

Need intuition - how do they simplify the Q function for gaussian mixture EM?

asked Mar 3, 2017 at 16:01

2 votes

1 answer

285 views

why isn't the tscv function allowing for step-size other than 1?

asked Oct 15, 2019 at 12:01

2 votes

0 answers

23 views

imbalanced regression problem + lower bound prediction + custom error weighting

asked Jan 28, 2020 at 9:16

2 votes

0 answers

35 views

Is there a relation between the characteristic roots and the s.e of the arima coefficients?

asked Nov 3, 2019 at 8:37

2 votes

1 answer

2k views

catboost does not overfit - how is that possible?

asked Feb 23, 2020 at 14:30

2 votes

1 answer

28 views

when fitting a dynamic regression model to a TS, what would happen if we first fit a regression model and then fit an ARMA?

asked Nov 13, 2019 at 12:10

2 votes

1 answer

388 views

BERT masking - why does it require sampling, and how does it mitigate the mismatch of the [MASK] token when fine-tuning?

asked Jul 2, 2020 at 15:34

1 vote

0 answers

24 views

alias method - for n discrete values, should the table size be n-1 or n? [closed]

asked May 30, 2020 at 7:05

Stack Exchange Network

52 Questions

What is the proper usage of scale_pos_weight in xgboost for imbalanced datasets?

Why is P(A,B|C)/P(B|C) = P(A|B,C)?

Can you give a simple intuitive explanation of IRLS method to find the MLE of a GLM?

t.test returns an error "data are essentially constant"

Why is the variable importance metric suggested by Breiman specific only to random forests?

Can you give an intuition behind the FTRL update step?

Fisher's score function has mean zero - what does that even mean?

can you explicitly show me the first iteration of newton-raphson and fisher scoring?

need intuition: why is there no winning strategy to this game?

Why is it that when the score variance gets LARGER we get MORE confident about the MLE estimate?

In GLM do we try to model E(T(y)) or E(y)?

Should I avoid mixed ARMA models?

Feature importance for xgboost classification of a sample

when fitting a regression model to a time-series, can I use lagged values of the time-series itself?

using the test population as an eval_set when doing hyperparameter optimization

Why are these 2 ARIMA formulations equivalent?

Can we do hypothesis testing for OLS estimator without assuming normality for the errors?

Is it necessary to use warm_start when tracking oob_score in scikit RandomForestClassifier?

Interpreting random survival forest mortality

how come that the variance of an estimate is smaller than the variance of the true value?

time series forecasting - predicting the next 24 hours

LIME - local linear explanations for global black-box models work for LogisticRegression but not for RandomForest?

Need intuition - how do they simplify the Q function for gaussian mixture EM?

why isn't the tscv function allowing for step-size other than 1?

imbalanced regression problem + lower bound prediction + custom error weighting

Is there a relation between the characteristic roots and the s.e of the arima coefficients?

catboost does not overfit - how is that possible?

when fitting a dynamic regression model to a TS, what would happen if we first fit a regression model and then fit an ARMA?

BERT masking - why does it require sampling, and how does it mitigate the mismatch of the [MASK] token when fine-tuning?

alias method - for n discrete values, should the table size be n-1 or n? [closed]