class Partition(HeavyTransformer):
"""Class to perform, e.g. Expand+kfoldCV.
This task is already done by function split(),
but if performance becomes a concern, this less modular solution is a
good choice.
TODO: the current implementation is just an alias for the nonoptimized
previous solution.
"""
def __init__(self,
split_type='cv',
partitions=10,
test_size=0.3,
seed=0,
fields=None):
if fields is None:
fields = ['X', 'Y']
config = self._to_config(locals())
# config cleaning.
if split_type == "cv":
del config['test_size']
elif split_type == "loo":
del config['partitions']
del config['partition']
del config['test']
del config['seed']
elif split_type == 'holdout':
pass
else:
raise Exception('Wrong split_type: ', split_type)
super().__init__(config)
from pjml.macro import split
self.transformer = Chain(
Expand(), split(split_type, partitions, test_size, seed, fields))
def _apply_impl(self, data):
splitter_model = self.transformer.apply(data)
applied = splitter_model.data.last_transformations_replaced(
drop=self.transformer.size,
transformation=self.transformations('a')[0])
return Model(self, data, applied, splitter_model=splitter_model)
def _use_impl(self, data, splitter_model=None):
used = splitter_model.use(data)
return used.last_transformations_replaced(
drop=self.transformer.size,
transformation=self.transformations('u')[0])
@classmethod
def _cs_impl(cls):
raise NotImplementedError
OnlyUse(Calc(functions=['flatten'])),
OnlyUse(Report('flatten S ... S: $S')),
OnlyUse(Calc(functions=['mean'])),
OnlyUse(Report('mean S ... S: $S')),
Report('End ...\n'),
)
# diversidade,
# Lambda(function='$R[0][0] * $R[0][1]', field='r')
print('sample .................')
pipe = full(rnd(expr, n=5), field='S', n=1).sample()
#
# pipes = rnd(expr, n=5)
#
# magia = Multi(pipes) -> Diversity() -> Agrega()
# magia.apply()
# coll = magia.use()
#
# pipe = full(pipes, field='S', n=1).sample()
print('apply .................')
data = Workflow(File("abalone3.arff"), Binarize()).apply().data
c = Chain(pipe.wrapped, Report())
model = c.apply(data)
print('use .................')
dataout = model.use(data)
)
# {history.last.config['function']}
print(expr)
print('sample .................')
pipe = full(rnd(expr, n=10), field='S').sample()
pipe.enable_pretty_printing()
print(f'Pipe:\n{pipe}')
print(f'Wrapped:\n{pipe.unwrap}')
pipe = Chain(File('abalone3.arff'), Binarize(), Split(), pipe.unwrap,
Metric(), Report())
print('apply .................')
model = pipe.apply()
# print(222222222222222, dataout.history)
# data morre no apply() do predictor
print('use .................')
# print(3333333333333333, dataout.history)
# RUS desaparece no use()
exit(0)
#
# # AutoML ===================================================================
