Parallel implementations of Logistic Regression.

Maven modules:

1. logreg-common: Common functionality regarding Machine Learning and Logistic Regression. Also contains some sequential implementations.

2. logreg-mapred: Hadoop Jobs for Logistic Regression

• Ensemble + SGD (mahout) Training
• Iterative Batch Gradient descent training
• Iterative L-BFGS training (uses Parallel Gradient-computation job)
• Forward Feature Selection using SFO (Single Feature optimization)

3. logreg-pact: Stratosphere Ozone Jobs for Logistic Regression

4. aim3-logreg-rcv1: Flexible preprocessing for rcv1-v2 (base on cuttlefish)

• Nice CLI
• contains v2 patch
• flexible time based splits
• multiple tf-idf weighting options
• ...

Developed for DIMA group at TU Berlin www.dima.tu-berlin.de