Reputation
12,913
Top tag
Next privilege 15,000 Rep.
Protect questions
Badges
7 62 115
Impact
~1.0m people reached

14h
revised Dimensionality reduction (SVD or PCA) on a large, sparse matrix
added 225 characters in body
Apr
28
comment Methods & CRAN packages to predict probability using neural networks or others machine learning algorithms
@HaskellFun It does take a test set. see ?predict.train. I updated my answer to demonstrate.
Apr
28
revised Methods & CRAN packages to predict probability using neural networks or others machine learning algorithms
added 6 characters in body
Apr
23
awarded  Famous Question
Apr
18
comment Using browser version types/numbers in Analysis in R
@Jessica Really like your encoding. I'm going to use that next time I'm analyzing browser data!
Apr
18
comment Python machine learning brute force
@Moebius Glad to hear that!
Apr
18
comment Python machine learning brute force
@Scortchi Totally correct! It's a tricky problem.
Apr
15
answered Python machine learning brute force
Apr
15
comment R Caret - Repeated CrossValidation, finalModel and ROC curves
@ThomasJvr Sounds like a case of too much documentation! =D. You're using a glm, so that methodology doesn't really apply. A glm is a glm— there are no parameters to tune. However, for OTHER models (e.g a random forest) there are parameters that control the model fit (e.g. mtry). caret tunes the parameters during cross-validation using "grid search," and then fits the final model using the best parameters from the grid search. Since a glm has no parameters, this doesn't really apply, but caret was really designed for models like random forests or gbms that have tunable parameters.
Apr
15
awarded  Popular Question
Apr
14
comment R Caret - Repeated CrossValidation, finalModel and ROC curves
@ThomasJvr If you'd bothered to read ?train, you could have figured this out on your own: "The combination with the optimal resampling statistic is chosen as the final model and the entire training set is used to fit a final model. "
Apr
12
awarded  Revival
Apr
12
comment How to handle over-prediction in Random-Forest
Once approach would be to split your data into a training set, calibration set, and test set. Fit your random forest on the training set, use the predictions on the calibration set + a linear regression model to learn an adjustment to the random forest, and use both models to predict on the test set.
Apr
12
answered R Caret - Repeated CrossValidation, finalModel and ROC curves
Apr
5
comment Using browser version types/numbers in Analysis in R
Then having 500 dummy variables is probably fine. You might want to drop any levels with <5 observations (or maybe 10).
Apr
5
comment Using browser version types/numbers in Analysis in R
How many observations do you have? You could make hierarchical variables (.e.g example <- c("IE", "IE-10") and example2 <- c("Chrome", "Chrome-47")) and throw them into something like glmnet for variable selection.
Apr
4
comment R libraries for deep learning
@Patric I can't see the link
Apr
4
comment R libraries for deep learning
@Patric What's the preload trick?
Mar
30
awarded  Popular Question
Mar
12
awarded  Revival