Experiment with alternative distance metrics to WMD

[marco-c]

Currently, after we transform the stack trace to a vector, we're using WMD (http://jmlr.org/proceedings/papers/v37/kusnerb15.pdf) to calculate distances.
We could experiment with alternative distance metrics that might be more suitable to stack traces. E.g. the position of the words in the vector is quite important for us; the lowest words in the vector (functions closer to the crash point) are likely more important for the clustering than highest words (functions closer to main).

[whatust]

I started by implementing the in-line wmd code as was suggested by the TODO on the code commentary, however I need some instructions on how to commit the changes I've made.

[marco-c]

@whatust after you fork this repo, clone it on your machine.
Then do, in the directory where you cloned it:

git remote add upstream https://github.com/marco-c/crashsimilarity.git

When you want to work on something:

# If the branch doesn't exist yet:
git checkout -b YOUR_FEATURE_NAME
# If the branch already exists:
git checkout YOUR_FEATURE_NAME

# DO YOUR CHANGES

git status
git add LIST_OF_FILES_THAT_YOU_MODIFIED
git commit -m "Description of what you have done"
git push origin feature_name

When you want to switch to working on something else, after you add and commit the changes on the current branch, simply do git checkout YOUR_FEATURE_NAME_2.

When your pull request is merged:

git checkout master
git pull upstream master

Now, you can open https://github.com/marco-c/crashsimilarity in your browser, and GitHub will ask you to submit a pull request.