Switch branches/tags
Nothing to show
Find file History
Latest commit 73892ae Aug 15, 2018
Permalink
..
Failed to load latest commit information.
data save Feb 10, 2017
dump fixed draw.py Jun 16, 2017
fault_prediction save Feb 10, 2017
fig save Feb 15, 2017
figures save Feb 10, 2017
tests save Jan 10, 2018
without_smote Update test_without.py Aug 15, 2018
README.md save Feb 1, 2017
__init__.py save Feb 1, 2017
setup.py save Feb 1, 2017

README.md

Fault Prediction -- pre-release name

The following are steps to install.

  • Install virtual environment. (Optional but recommended)
  • Then run the following:
  $ cd fault_prediction # Basically set cwd to fault_prediction
  $ pip install -e .
  $ python -m tests.test _test filename
  • filename just "ant"
  • Magic

Experiments:

  • 6 Learners [KNN, NB, SVM, LR, DT, RF]
  • with smote and without smote, 5x5 cross val
  • only 14 datasets, other datasets from openscience were very small.
  • compared against 4 measures, [Precision, recall, accuracy, f_score]

Conclusions:

  • Considering all the measure almost all learners performed the best. But in most cases we saw, Random forest to win.
  • Smote is needed for highly imbalanced classes.
  • Runtimes are within 5-20 mins for each dataset with 6 learners repeating 25 times. Only 2 datasets took about 2hours each.

Results:

With Smote:

file

file

file

file

Without Smote:

file

file

file

file