Gradient boosting machines are generally very slow in implementation because of sequential model training.Thus, #XGBoost is focused on computational speed and model performance.
#LightGBM beats all the other algorithms when the dataset is extremely large. LightGBM is a gradient boosting framework that uses tree-based algorithms and follows #leafwise approach while other algorithms work in a #levelwise approach pattern.
#Boosting in general decreases the bias error and builds strong predictive models. #Boosting has shown better predictive accuracy than #bagging, but it also tends to over-fit the training data as well #ensemblemodelling
#Earlystopping is when you are training your dataset but do not go all the way till the point you cost function is the least so as to keep the weights mid sized.The more you train,the bigger the weights of your model become hence it simulateneously gives #regularization effect
There are various methods to calculate the similarity between two objects while building a #recommendationsystem. Distance scores like #euclidean, #cosine, #pearson#corelation can be used, there is no good or bad. decide using all three and see which gives better results
#Recommendor systems are widely divided into #contentbased and #collaborativefiltering. For former, you are aware of different features vectors that define the product to be recommended while in latter, it is an incremental process of learning product as well as user features
Some info on activation functions. #sigmoid to convert outputs between 0 to 1, #tanh for -1 to 1, #relu where anything less thn zero has output of 0, #softmax again to collapse outputs between 0 to 1 - used in #multilable#classification