*** Dubbing: [ English ] [ 한국어 ]
In this video, we will analyze the Exact Greedy Algorithm of XGBoost classification in detail.
In the last video, we looked at the training process and the prediction process for XGBoost classification through a simple example. In this video, we will derive the formulas used in these processes and analyze in detail the algorithm presented in the paper.
Let's take a look at the algorithm introduced in the XGBoost paper.
Section 2.1 of this paper sets regularized learning objective function and optimizes this objective function to create the formulas that calculate the score, gain, and the output value used in the previous video. In regression the mean-squared error is used for this loss function.
In classification, binary cross entropy is used as the loss function. By simply changing the loss function, this algorithm can work as both regression and classification. Optimizing the objective function provides a formula for calculating the output value of each leaf node.
Let's analyze this in detail.
#ExtremeGradientBoosting #XGBoost #XGBoostClassification #ExactGreedyAlgorithmForSplitFinding
