def test_translate_alone():
feature_set = sorted(["A", "B"])
for l in [32, 64, 128]:
feature_df_list = [
pd.DataFrame({
"time": pd.to_datetime(list(range(l)), unit="s"),
"A": np.array(list(range(l))),
"B": np.array(list(range(l))),
"y": np.ones(l)
}) for _ in range(1)
]
for (look_back, look_forward) in [(3, 2), (1, 0)]:
custom_transforms = list()
custom_transforms.append(remove_false_anchors_factory("y"))
custom_transforms.append(
split_flat_df_by_time_factory(look_back, look_forward, 1))
translate = Translate(features=feature_set,
look_back=look_back,
look_forward=look_forward,
n_seconds=1,
custom_transforms=custom_transforms,
normalize=False)
X, y = translate.scale_and_transform_session(feature_df_list[0])
tools.eq_(X.shape, (l - (look_back + look_forward),
(look_back + look_forward + 1), 2))
tools.eq_(len(y), l - (look_back + look_forward))
# first elements should slide forward in time one element at a time
np.array_equal(X[:, 0, 0], np.array(list(range(l))))
# second elements should slide forward one at a time starting at 1
np.array_equal(X[:, 1, 0], np.array(list(range(l))) + 1)
def test_time_gaps_split():
n = 50
df = pd.DataFrame({
"time": range(n),
"A": np.random.randn(n),
"B": np.random.randn(n),
"y": np.random.randint(2, size=n)
})
df.loc[25:, "time"] += 1
df["time"] = pd.to_datetime(df["time"], unit="s")
sfd = split_flat_df_by_time_factory(look_back=5,
look_forward=5,
n_seconds=1)
res = sfd(df)
tools.eq_(len(res), 2)
assert df.equals(pd.concat(res, axis=0))
def test_time_gaps_fully_segmented():
n = 50
gap = 2
df = pd.DataFrame({
"time":
pd.to_datetime(list(range(0, n * gap, gap)), unit="s"),
"A":
np.random.randn(n),
"B":
np.random.randn(n),
"y":
np.random.randint(2, size=n)
})
sfd = split_flat_df_by_time_factory(look_back=0,
look_forward=0,
n_seconds=1)
res = sfd(df)
tools.eq_(len(res), n)
def test_time_gaps():
n = 50
for gap in range(1, 3):
df = pd.DataFrame({
"time":
pd.to_datetime(list(range(0, n * gap, gap)), unit="s"),
"A":
np.random.randn(n),
"B":
np.random.randn(n),
"y":
np.random.randint(2, size=n)
})
sfd = split_flat_df_by_time_factory(look_back=5,
look_forward=5,
n_seconds=gap)
res = sfd(df)
assert df.equals(res[0])
tools.eq_(len(res), 1)