Planned Modules

Below is a list of planned modules, in no particular order, with some general notes for each. A checked box indicates that the first draft of the module has been committed (not that it's fully complete!).

• Classes

  • Unified predictor interface (a la sklearn fit/ predict/ score), see docs

• Metrics

  • general measures of model quality for classifiers and regressions
  • include cluster quality measures?
  • WIP: needs to be cleaned up
  • parallel confusion matrix?
  • k-fold cv

• Naive Bayes

  • still needs some work, but initial draft is committed

• Clustering

  • encoding for categorical variables?

• Decision Tree

  • information gain/ entropy measures
  • selection of pruning techniques
  • make it modular, like kmeans, to mix and match techniques
  • implement pruning methods

• Ensemble Methods

  • Ada boost
  • random forest (better in tree module?)

• SVM

  • kernel tricks

• KNN

• Neural Net

• Linear Regression

  • gradient descent

• Frequent Pattern Mining

  • FP growth
  • frequent sequence

• Utilities

  • data loading and IO, data generation
  • parallel strategies & pipelining

---

In addition to implementing these modules, I also plan to create a more portable benchmark suite that encapsulates the generation or loading of data. The benchmarks will be organized in a way that will make comparison to analogous methods in other languages (e.g. Python) will be easy to make.