understanding the problem

• outcome/data are already selected for you
• what is important to capture in a CDR for this problem?

understanding the data

what features are reliable/useful/predictive?

modeling

writeup

• checking stability and judgement-calls
• comparison with baseline model

fork the repo

make fork private + include your teammates

why are we using python 🐍?

python is better

R to python

indentation matters!

setting up python

• python --version should give 3.6 or higher (might need to type python3)

• easier if you install things by making a venv

python3 -m venv rule-env  // create the env
source rule-env/bin/activate  // activate the env

installation

git clone https://github.com/Yu-Group/rule-vetting  // clone the repo
cd rule-vetting
pip install -e .

editing in python

you can use any editor, maybe jupyterlab or pycharm

pandas

numpy

seaborn / matplotlib

sns.lmplot(x="x", y="y", col="dataset", hue="dataset", data=df,
           col_wrap=2, ci=None, palette="muted", height=4,
           scatter_kws={"s": 50, "alpha": 1})

scikit-learn

scikit-learn example

from sklearn import tree
X = [[0, 0], [1, 1]]
Y = [0, 1]
clf = tree.DecisionTreeClassifier()
clf = clf.fit(X, Y)
clf.predict([[2., 2.]])

🔎 imodels

imodels usage

from imodels import RuleFitClassifier
model = RuleFitClassifier()
model.fit(X_train, y_train)
preds = model.predict(X_test)
preds_proba = model.predict_proba(X_test)
print(model)

package for facilitating PCS analysis, especially stability

vflow usage

• implemented in the dataset template - will help speed things up for you

quick start

• look at the example project (iai_pecarn)
• look at the template files
• probably just start with some notebooks and then implement template functions

understanding the template

specifying judgement calls

ex function with a judgement call

tests

pytest --project <your_project_name>

e.g. pytest --project iai_pecarn

ex notebooks