Porting `dplyr` to python: Verb argument evaluation
Function calls as verb arguments
What if we want to do:
iris >> mutate(mean_sepal_length=mean(f.Sepal_Length))
f.Sepal_Length and f.Petal_Length refer to two columns in iris. We need to defer the evaluation of the arguments, too, as we did for the verbs.
To retrieve the series in the data, we need to let f.Sepal_Length to be evaluated when a data frame is available:
class Symbolic:
def __getattr__(self, name):
self.name = name
return self
def _eval(self, data):
return data[self.name]
Similarly, when python sees mean(f.Sepal_Length), it gets evaluated. But python hasn't seen the data yet. So the expression needs to be evaluated later, inside Verb.__rrshift__(). This needs us to turn the function mean into Verb-like, but not exactly a Verb, because a verb takes data as the first argument but the function may or may not.
Ask the verb to evaluate the arguments with the data:
class Verb:
# ...
# Evaluate args and kwargs in verb
def __rrshift__(self, data):
args = eval_args(args, data)
kwargs = eval_kwargs(kwargs, data)
return self.func(data, *args, **kwargs)
eval_args = lambda args, data: (
arg._eval(data) if isinstance(arg, (Symbolic, Func)) else arg
for arg in args
)
eval_kwargs = lambda kwargs, data: {
key: (val._eval(data) if isinstance(val, (Symbolic, Func)) else arg)
for key, val in kwargs.items()
}
Define the function:
class Func(Verb):
def _eval(self, data):
# evaluate my args/kwargs too
args = eval_args(args, data)
kwargs = eval_kwargs(kwargs, data)
return self.func(*args, **kwargs)
@Func
def mean(series):
return series.mean()
Operators as verb arguments
Operators in python are actually functions, too. For example:
import operator
a = b = 1
a + b # 2
operator.add(a, b) # 2
Then we are able to turn the operators into function in:
class Symbolic:
def __add__(self, other):
return Func(operator.add)(self, other)