def printing_test(arq="iris.arff"):
print(Chain(Map(select(File(arq)))))
exp = Workflow(
File(arq),
Partition(),
Map(PCA(), SVMC(), Metric(enhance=False)),
Map(Report("<---------------------- fold"), enhance=False),
Summ(function="mean", enhance=False),
Reduce(),
Report("mean ... S: $S", enhance=False),
)
print(exp)
print(select(DT(), SVMC()))
sel = select(DT(), SVMC())
print(sel)
print(Map(DT()))
exp = ChainCS(
File(arq),
Partition(),
Map(PCA(), select(SVMC(), DT(criterion="gini")),
Metric(enhance=False)),
Report("teste"),
Map(Report("<---------------------- fold")),
)
print(exp)
def ger_workflow(seed=0, arq="iris.arff"):
np.random.seed(seed)
workflow = Workflow(File(arq),
Partition(),
Map(PCA(), select(SVMC(), DT(criterion="gini")),
Metric(enhance=False)),
Summ(function="mean", enhance=False),
Reduce(),
Report("Mean S: $S", enhance=False),
seed=seed)
return workflow
def test_with_summ_reduce(arq="iris.arff"):
pipe = Workflow(
File(arq),
Partition(),
Map(PCA(), SVMC(), Metric()),
Map(Report("<---------------------- etapa")),
Summ(),
Reduce(),
Report("mean ... S: $S"),
)
train, test = pipe.dual_transform()
print("Train..............\n", train.history ^ "longname")
print("Test..........\n", test.history ^ "longname")
def test_split_train_test(arq="iris.arff"):
pipe = Cache(
File(arq),
TsSplit(
), # TsSplit should come before TrSplit to ensure the same original data is used as input for both.
TrSplit(),
PCA(),
SVMC(),
Metric(enhance=False),
Report("metric ... R: $R", enhance=False),
storage_alias="oka")
train, test = pipe.dual_transform()
print("Train..............\n", train)
print("Test..........\n", test)
def test_partition(arq="iris.arff"):
pipe = Workflow(
File(arq),
Partition(),
Map(PCA(), SVMC(), Metric(enhance=False)),
Summ(function="mean", enhance=False),
Reduce(),
Report("mean ... S: $S", enhance=False),
Report("$X"),
Report("$y"),
)
train, test = pipe.dual_transform()
print("Train..............\n", train)
print("Test..........\n", test)
def test_check_architecture2(arq="iris.arff"):
pipe = Workflow(
File(arq),
Partition(),
Map(PCA(), SVMC(), Metric(enhance=False)),
Summ(field="Y", function="mean", enhance=False),
Report("mean ... S: $S", enhance=False),
)
# tenho file na frente
train_ = pipe.enhancer.transform(sd.NoData)
test_ = pipe.model(sd.NoData).transform(sd.NoData)
test_ = pipe.model(sd.NoData).transform((sd.NoData, sd.NoData))
train_, test_ = pipe.dual_transform(sd.NoData, sd.NoData)
train_, test_ = pipe.dual_transform(sd.NoData, (sd.NoData, sd.NoData))
def random_search(arq="iris.arff"):
np.random.seed(0)
exp = Workflow(
File(arq),
Partition(),
Map(PCA(), select(SVMC(), DT(criterion="gini")), Metric()),
# Map(Report("<---------------------- fold"), enhance=False),
Summ(function="mean"),
Reduce(),
Report("Mean S: $S"),
)
expr = sample(exp, n=10)
result = optimize(expr, n=5)
result.disable_pretty_printing()
print(result)
def test_check_architecture(arq="iris.arff"):
pipe = Workflow(
File(arq),
Partition(partitions=2),
Map(PCA(), SVMC(), Metric(enhance=False)),
Summ(field="Y", function="mean", enhance=False),
)
# tenho file na frente
train_01 = pipe.enhancer.transform(sd.NoData)
test_01 = pipe.model(sd.NoData).transform(sd.NoData)
train_02, test_02 = pipe.dual_transform(sd.NoData, sd.NoData)
# Collection uuid depends on data, which depends on consumption.
for t, *_ in train_01, train_02, test_01, test_02:
# print(111111111, t.y)
pass
assert train_01.uuid == train_02.uuid
assert test_01.uuid == test_02.uuid
def test_sequence_of_classifiers(arq="abalone.arff"):
pipe = Workflow(
File(arq),
Binarize(),
Report('1 {X.shape} {history^name}'),
PCA(n=5),
SVMC(),
Metric(),
Report('2 {X.shape} {history^name}'),
DT(),
Metric(),
Report('3 {X.shape} {history^name}'),
)
print('Enh')
train = pipe.enhancer.transform(sd.NoData)
print('Mod')
test = pipe.model(sd.NoData).transform(
sd.NoData) # TODO: pq report não aparece no test?
print()
print("[test_sequence_of_classifiers] Train.........\n",
train.history ^ "longname")
print("[test_sequence_of_classifiers] Test..........\n",
test.history ^ "longname")
exit(0)
Save('lixo.arff').apply(dout)
# ML 2 ========================================================================
pipe = Pipeline(
File('iris.arff'),
OverS(sampling_strategy='not minority'),
ApplyUsing(NB('bernoulli')),
Metric(functions='accuracy'),
# Report('Accuracy: $r {history}'),
Report('Accuracy: $r'),
ApplyUsing(DT(max_depth=2)),
Metric(functions='accuracy'),
Report('Accuracy: $r'),
ApplyUsing(SVMC(kernel='linear')),
Metric(functions='accuracy'),
Report('Accuracy: $r'),
)
m = pipe.apply()
dataout2 = m.use()
# ML 3 ========================================================================
pipe = Pipeline(
File('iris.arff'), Cache(MFE()),
Report('\nMeta-features Names: $Md \nMeta-features Values: $M \n {name}'))
dataout = pipe.apply().data
"""
Problemas filosoficos
obs. Containers sempre contêm referências a outros transformers (sejam leves ou
def test_pca(arq="iris.arff"):
cs = File(arq).cs
pipe = Workflow(File(arq), Split(), PCA(), SVMC(), Metric())
train, test = pipe.dual_transform()
print("Train..............\n", train.history ^ "name")
print("Test..........\n", test.history ^ "name")
def test_metric(arq="iris.arff"):
pipe = Workflow(File(arq), Split(), SVMC(), Metric(enhance=False))
train, test = pipe.dual_transform()
print("Train..............\n", train)
print("Test..........\n", test)
def test_split(arq="iris.arff"):
pipe = Workflow(File(arq), Split(), SVMC())
train, test = pipe.dual_transform()
print("Train..............\n", str(train))
print("Test..........\n", str(test))
def test_svmc(arq="iris.arff"):
cs = File(arq).cs
pipe = Workflow(File(arq), SVMC())
train, test = pipe.dual_transform()
print("Train..............\n", train)
print("Test..........\n", test)