Assembling models Shifu combo
Assembling models is try to leveraging different kinds of machine learning algorithm and combining them together to create an assemble model. See below:
Different machine learning algorithm has different capability to detect data pattern in large dataset. Assembling models could leverage all the capability of different algorithms and help to boost the overall performance. Meanwhile, assembling models can prevent overfit in some algorithm.
shifu combo -new NN,LR,GBT,LR
After this command a new file ComboTrain.json will be generated. You can edit configuration for each sub-models and assemble model.
shifu combo -init
Run init command, the folders for all sub models and assemble model will be created under current directory.
shifu combo -run
Train each sub models, use sub-models to evaluate training data set, and then use training data set evaluation result to train assemble models.
shifu combo -eval
Evaluate model performance for all sub-models and assemble models. You can check model performance in logs/shifu.log in each model directory.
