liblinear-experiment

This repo is the project of mpml, aiming to solve a two-class classification problem using liblinear in multi-kernal machines.

The min-max network is used to decompose the task so that it can be fully parallelized. I will campare the performance between randomized partition and label-based partition.

Todo

• [x] Using liblinear to solve the origin problem
• [x] Using min-max network to solve the same problem
• [x] Using pickle to dump/load data
• [x] Parallelize the min-max network
• [x] Partition data by labels
• [x] Measure the time cost in different solution
• [x] calculate f1 value
• [x] draw ROC graph
• [x] using dump/load in data file IO
• [x] Avoid reading data file for multiple times
• use DBMS to manage data
• [d] use queue to exchange data between processes
• [x] use logging module