[Source Code] Inquiry about Boosting Iteration Logic in XGBoost's Training Loop

#1

Why is the version variable incremented twice within the loop in the provided code snippet? As a result, version is always an even number at the beginning of each iteration? Then the number of boosting iterations is actually half of the specified param_.num_round?

// load in data.
...

// initialize the learner.
...

// start training.
int32_t version = 0;
for (int i = version / 2; i < param_.num_round; ++i) {
    if (version % 2 == 0) {
	    learner_->UpdateOneIter(i, dtrain);
	    version += 1;
	}
	...
    version += 1;
}

// save model
...

Why don’t we simply use the following logic? Why do we need the version variable? What’s it for?

// load in data.
...

// initialize the learner.
...

// start training.
for (int i = 0; i < param_.num_round; ++i) {
    learner_->UpdateOneIter(i, dtrain);
}

// save model
...

or

// load in data.
...

// initialize the learner.
...

// start training.
for (int i = 0; i < param_.num_round/2; ++i) {
    learner_->UpdateOneIter(i, dtrain);
}

// save model
...