Kaggle Utils
Here, I will store my utils scripts that helped me in the Kaggle competitions
AutoEnsembler
AutoEnsembler is a simple tool that allows to achieve +2%-3% validation accuracy during models ensembling.
It works by smartly splitting the validation set, ierarchically selecting best blending parameters and by consequently shrinking the grid size. It has n_jobs
parameter to allow for faster computations.
AutoEnsembler was used in these competitions:
- Humpback Whales Identification: 0.6% LB improvement, 2% CV improvement compared to linear blending.
FileIO
FileIO is a simple library that allows you to store numerical objects in a fast and efficient way.
Usage:
import numpy
import fileio
array = {
'numpy': numpy.array([1, 2]),
'string': 'fileio!'
}
fileio.save(array, 'example')
assert fileio.check('example')
assert fileio.load('example')['string'] == 'fileio!'
fileio.remove('example')