XGBoost

What do you like best about XGBoost?

The best thing about XGBoost is it provides parallel processing in the machine learning model development; with the help of 4 cores and parallel processing, i was able to develop a machine learning model on 30 Million subscribers in 2 hours.

What do you dislike about XGBoost?

What I don't like about XGBoost is it doesn't handle the outliars in the dataset while machine learning model development.

What problems is XGBoost solving and how is that benefiting you?

I am using XGBoost for the machine learning model development on a large dataset. It is speedy and provides good results in terms of accuracy and other matrices.

What do you like best about XGBoost?

It's the best performing stand-alone algorithm (not counting deep learning algorithms which is whole another field) famous for winning many online machine learning competitions. It runs fast and performs better than bagging algorithms because it learns from the mistakes of previous tree models that were built within it. It is possible to tune XGBoost for various metrics, too so if you want a high recall, you can do it with the help of GridSearchCV. It is very efficient compared to famous Random Forest algorithm.

What do you dislike about XGBoost?

That it is not a part of a bigger package such as Anaconda but we have to install it separately. Also, its greatness comes with the cost of overfitting just like deep neural networks. It learns so good that after hyperparameter tuning it overfits more than other algorithms.

Recommendations to others considering XGBoost:

Be careful when hyperparameter tuning, it can overfit even though train and test data were separated.

What problems is XGBoost solving and how is that benefiting you?

I am solving Machine Learning problems with XGBoost. It learns and performs very well both in terms of performance metrics such as accuracy and fscore as well as training time.

What do you like best about XGBoost?

I have used XGBoost models for many ML competition problems so far. Every time I could achieve a high accuracy and high performance model through using XGBoost. XGBoost is well known for its better performance and efficient memory management in ML community. Therefore, I highly recommend anyone who is new to the field to learn and use XGBoost. It is must to be in your ML toolkit.

What do you dislike about XGBoost?

The underlying concept of the algorithm is somewhat hard to understand at first. And the model has a large number of hyperparameters. Hence, at the beginning, it is difficult to understand the role each hyperparameter plays. But after some reading of the theory of the algorithm etc. the model becomes easy to comprehend and use.

Recommendations to others considering XGBoost:

Take some time to understand and its concept. Then, it becomes very easy to use XGBoost.

What problems is XGBoost solving and how is that benefiting you?

I used XGBoost to train models for financial data mostly. But, apart from business problems, I could use the models in ML and data science competitions, I recently participated. Each and every instance, I could achieve high prediction performances from XGBoost. The ability to give high accuracy models is one of the main advantages in XGBoost. Apart from it, it is fast in execution and easy to use. Hence, I believe this algorithm is one of the best tools a data scientist or a ML engineer should possess.

What do you like best about XGBoost?

- XgBoost is a type of library which you can install on your machine. C++, Java, Python with Sci-kit learn and many more.

- It does parallelization tree construction using all CPU cores

- The implementation of the algorithm was engineered for the efficiency of computing time and memory resources.

- Xgboost ensures the execution speed and model performance

- XGBoost internally has parameters for cross-validation, regularization, user-defined objective functions, missing values etc

- It helps to reduce overfitting.

What do you dislike about XGBoost?

There is nothing much I dislike about the Xgboost but for me sometimes tuning the parameters is bit hectic.

Recommendations to others considering XGBoost:

The main reasons to use XgBoost is its execution speed and increase in model performance. You can solve the classification, regression and ranking problem very easily. There are lot of parameters and user define functions which will surely come in handy. It is always better to try XgBoost first.

What problems is XGBoost solving and how is that benefiting you?

If you don't know what to do use XgBoost algorithm. I will suggest using XgBoost according to your requirements. I use XgBoost over light GBM because it gives me better results. Mostly in Kaggle competitions, I use such algorithms.

What do you like best about XGBoost?

Have used XGBoost multiple times, and it is a very intuitive library that is easy to pick up quickly for the task I had at hand (fairly straightforward gradient boosting task). I only used the package in R form, but have heard good things from colleagues who much more regularly use gradient boosting for predictive projects; XGBoost seems to be the go-to library for boosting for multiple Data Scientists that I work with.

What do you dislike about XGBoost?

Nothing comes to mind; it is an efficient and easy to use gradient boosting framework. The support for the R version seems a little less than the Python version, but the R version performed well for my needs (relatively small dataset, no multicore processing or need for intense parallelization.

What problems is XGBoost solving and how is that benefiting you?

We leverage machine learning platforms and statistical methods to evaluate promising social interventions, pulling from many different fields and frameworks. XGBoost is a good implementation of a key technique in machine learning, and it was relatively simple to implement in a public policy context.