ActiveLearningDemo

How to run

1. step one

put the dataset folder and use command below to split the dataset to the required structure

run utils.py 

For each dataset, six .mat documents should be included: TrainingMatrix.mat, TrainingLabels.mat, TestingMatrix.mat, TestingLabels.mat, UnlabeledMatrix.mat and UnlabeledLabels.mat.

2. step two

Train the model. You can set arguments:

Active learning

optional arguments:
  -h, --help            show this help message and exit
  --src SRC             dataset path
  --dst DST             destination path
  --type TYPE           sample strategy:random, entropy, combine
  --solver SOLVER       model solver
  --max_iter MAX_ITER   max iteration of each training
  --k K                 samele added for each iteration
  --n N                 number of iterations
  --plot_type PLOT_TYPE
                        plot single for one case(single) or plot average for
                        entire database(average) 

You can utilize both one dataset with multiple subsets inside and one case of a dataset with only six .mat documents. By default, I used "newton-cg" solver and "combine" type which can train model with both strategies at once. To get results on different datasets directly, you can use:

python main.py --src your dataset path(./datasets/MMI) --dst output path(./img)

Result

1. MMI dataset

use "lbfgs" solver:

use "newton-cg" solver:

2. MindReading dataset

use "lbfgs" solver:

use "newton-cg" solver: