Models-Pipeline-Machine-Learning

Pipeline for machine learning binary problems, wrote completely with numpy.

Pipeline.py

• Pipeline: Class used to create a pipe:
  • setStages(): set an array of PipelineStage
  • addStages(): add an array of PipelineStage
  • fit(): get DTR (nFeature, nSample) and LTR (nSample,), it will return a model. It call iteratively compute() of a PipelineStage
• PipelineStage: Interface for a stage of a Pipeline
  • compute(): main method to start the stage
• VoidStage: A stage that doesn't do anything
• Model: Interface of a model
  • transform(): get DTE (nFeature, nSample) and LTE (nSample,), it will return the scores. It call iteratively compute() of a PipelineStage of preprocessing stages.
• CrossValidator: Class used to do CrossValidation
  • setEstimator(): set a Pipeline
  • setNumFolds(): set the number of Folds
  • fit(): get DTR and LTR, and it creates k-folds randomly from them, then it calls for each fold fit() of the Pipeline. At the end it returns the scores of the DTR.

classifiers.py

• MVG (PipelineStage): Multivariate Gaussian
• NaiveBayesMVG (PipelineStage): Multivariate Gaussian (Diag)
• TiedMVG (PipelineStage): Multivariate Gaussian (Single Cov)
  • setPiT(): set the re-balancing factor
• TiedNaiveBayesMVG (PipelineStage): Multivariate Gaussian (Diag Single Cov)
  • setPiT(): set the re-balancing factor
• LogisticRegression (PipelineStage): Logistic Regression
  • setLambda(): set the Lambda factor, it is the Regulizer Factor, 0 := Overfitting, 1 := Underfitting
  • setPiT(): set the re-balancing factor
  • setExpanded(): set True or False, if you want to use the Quadratic or Linear Model
• SVM (PipelineStage): Support Vector Machine
  • setK(): set the K factor, usually 1
  • setC(): set the C factor, 0 := Big Margin, 1 := Small Margin
  • setPiT(): set the re-balancing factor
  • setPolyKernel(): use a Polynmial Kernel, set c factor and d (degree)
  • setRBFKernel(): use a RBG Kernel, set Gamma factor
  • setNoKern(): no Kernel
• GMM (PipelineStage): Gaussian Mixture Model Clustering model used for Classification
  • setDiagonal(): use Diagonal Matrices of GMM density
  • setTied(): use same matrice for all components of GMM
  • setIterationLBG(): set the number of iteration of LBG, the final number of component is a power of 2 of the iteration
  • setAlpha(): set alpha factor, for LBG algorithm, it's the rescaling factor for the new starting points of the components at each iteration of LBG
  • setPsi(): set psi factor, used to avoid the problem of generative solutions, it's a limitation on the variation of the covatiance matrices of a GMM component

models.py

All the Class here implement the Model Interface, the main method is tranform()

preProc.py

All the pre-processing stage, are PipelineStage, the main method is compute()

• PCA
• LDA
• ZNorm
• L2Norm
• Gaussianization

Tools.py

Lots of usefull tools, used inside the whole project