# Tag Info

0

For Q1, check out these two stats.SE posts: Why does XGBoost have a learning rate? XGBoost Loss function Approximation With Taylor Expansion They bring up questions about the goodness of the second-order expansion, which is even more strict than asking about radius of convergence, and I think the discussion probably answers this reasonably well. I suspect ...

0

Many alleged "classification" models actually predict probabilities, and then there is some decision function that maps the probability to a category. The common decision function is to take the category that has the highest probability, but you can pick any threshold. You might even choose not to use a threshold and to do direct evaluation of the ...

1

I'm not sure that these terminologies are universal, but the xgboost documentation appears to be considering a "decision tree" to specifically mean that the predictions made are hard class predictions (the mode of the classes among training data in a leaf), not probability predictions, and therefore not usable for regression tasks. Regression trees ...

0

I agree with ‘Carlos Mougan,’ and add the comparison with these models. Basically, GB(Breiman 1996, 1999)、SGB(Friedman 2002) and XGB(Chen and Guestrin 2016) do sampling w/o replacement. GB uses the full sample set for each iteration. Thus, GB performs like sampling w/o replacement. So, the three use sampling w/o replacement by tradition. However, bagging and ...

0

Decision tree will split your data based on the most relevant features, it is not neccessary to give each decision tree with different feature. Let say an example, You have 3 features named as gender,profession and fare and the output required is some column. The most relevent features can be captured by different approaches, here i taking entropy and ...

1

You can convert DMatrix to NumPy array using dmatrix2np: from dmatrix2np import dmatrix_to_numpy converted_np_array = dmatrix_to_numpy(dmatrix) It's open-source, you can see its code here.

0

I dug into this a bit over at this answer, including a Colab notebook. I'll mostly refer you to that answer for the references, but I'll summarize here: At some point in time, xgboost had different behavior for low-cardinality discrete features, splitting at the actual data values; whereas continuous features have always(?) split at midpoints between ...

Top 50 recent answers are included