MNIST Active Learning
Requirements to install ( python2 or python3 )
python -m pip install -r requirements.txt --user --no-cache or
python3 -m pip install -r requirements.txt --user --no-cache
To run
python activemnist.py 1 10 2000 True old True 5 or
python3 activemnist.py 1 10 2000 True old True 5
Additional Running Info
######runs a test with the following parameters:
###1st parameter: number of runs (almost always want just 1 or will be extremely long)
###2nd parameter: how big each mini-batch should be
###3rd parameter: how many mini-batches in a run
###4th parameter: active learning turned on or not
###5th parameter: classification task ('old', '3s', 'split')
###6th parameter: re-samping turned on or not
###7th parameter: how often training is tested with test set
example: python activemnist.py 1 10 2000 True old True 5
would start one full run with mini-batches of size 10 for 2000 iterations with Active Learning
turned on, classifying digits from 0-9, with re-sampling and printing test results every 5 mini-batches
in addition it would then run 20 epochs at intervals of 250 label counts