Computational branches

If you have multiple forks, they should all have the same branches. Each branch will execute independently of the rest (non-fork inputs are shared between them). You can even create machine learning model forks, where a different model is applied on different branches:

from sklearn.linear_model import LogisticRegression, MLPClassifier

x, y = ...

classifier = fb.Fork(case1=LogisticRegression(), case2=MLPClassifier())
classifier = classifier.fit(x, y)
yhat = classifier.predict(x)

Forks automatically try to call wrapped class methods, i.e., classifier.fit is also a fork whose branches hold the outcome of applying fit on each branch's model. The inputs x,y could also have been forks, in which case each branch would have been trained on respective values.

Recall that branch values can be accessed via class fields, for example like yhat = (yhat.case1+yhat.case2)/2. This computation produces a factual value that is not bound to any branch. On the other hand yhat.case1 and yhat.case2 would be used during assessment of case1 sensitive attribute values and case2 sensitive attribute values.

print(classifier)
# case2: MLPClassifier()
# case1: LogisticRegression()
print(yhat)
# case2: [0 1 0 1 0 1 1 1]
# case1: [0 1 0 1 1 1 1 1]
print((yhat.case1+yhat.case2)/2)
# [0.  1.  0.  1.  0.5 1.  1.  1. ]

A visual view of how data are organized across branches follows. Some variables are identical but others obtain different values per branch. The same code is run on all branches concurrently and independently.