def test_truncation(truncation):
data = {"a": np.random.randn(2, 10), "b": np.random.randn(2, 10)}
duration = 10 if truncation is None else truncation
with pytest.warns(None) as w:
for i, (o, d) in enumerate(
utils.minibatch_generator(data,
2,
shuffle=False,
truncation=truncation)):
assert np.allclose(d["a"], data["a"][:, o:o + duration])
assert np.allclose(d["b"], data["b"][:, o:o + duration])
# pylint: disable=undefined-loop-variable
assert i == 10 // duration - (10 % duration == 0)
assert len(w) == (1 if truncation == 3 else 0)
# check truncation with n_steps input
for i, (o,
d) in enumerate(utils.minibatch_generator(10, None, truncation=3)):
assert o == i * 3
if i < 3:
assert d == 3
else:
assert d == 1
def test_minibatch_generator(shuffle):
inputs = {"a": np.arange(100)[:, None]}
targets = {"b": np.arange(1, 101)[:, None]}
x_all = []
y_all = []
for _, x, y in utils.minibatch_generator(inputs,
targets,
10,
shuffle=shuffle):
x_all += [x["a"]]
y_all += [y["b"]]
x_all = np.concatenate(x_all)
y_all = np.concatenate(y_all)
if shuffle:
assert not np.allclose(x_all, inputs["a"])
assert not np.allclose(y_all, targets["b"])
x_all = np.sort(x_all, axis=0)
y_all = np.sort(y_all, axis=0)
assert np.allclose(x_all, inputs["a"])
assert np.allclose(y_all, targets["b"])
x_all = []
y_all = []
with pytest.warns(UserWarning):
for _, x, y in utils.minibatch_generator(inputs,
targets,
12,
shuffle=shuffle):
assert x["a"].shape[0] == 12
assert y["b"].shape[0] == 12
x_all += [x["a"]]
y_all += [y["b"]]
x_all = np.sort(np.concatenate(x_all))
y_all = np.sort(np.concatenate(y_all))
assert len(x_all) == 96
assert len(y_all) == 96
if shuffle:
assert not np.allclose(x_all, np.arange(96)[:, None])
assert not np.allclose(y_all, np.arange(1, 97)[:, None])
else:
assert np.allclose(x_all, np.arange(96)[:, None])
assert np.allclose(y_all, np.arange(1, 97)[:, None])
def test_minibatch_generator(shuffle):
data = {"a": np.arange(100)[:, None], "b": np.arange(1, 101)[:, None]}
x_all = []
y_all = []
for _, d in utils.minibatch_generator(data, 10, shuffle=shuffle):
x_all += [d["a"]]
y_all += [d["b"]]
x_all = np.concatenate(x_all)
y_all = np.concatenate(y_all)
if shuffle:
assert not np.allclose(x_all, data["a"])
assert not np.allclose(y_all, data["b"])
x_all = np.sort(x_all, axis=0)
y_all = np.sort(y_all, axis=0)
assert np.allclose(x_all, data["a"])
assert np.allclose(y_all, data["b"])
x_all = []
y_all = []
with pytest.warns(UserWarning):
for _, d in utils.minibatch_generator(data, 12, shuffle=shuffle):
assert d["a"].shape[0] == 12
assert d["b"].shape[0] == 12
x_all += [d["a"]]
y_all += [d["b"]]
x_all = np.sort(np.concatenate(x_all))
y_all = np.sort(np.concatenate(y_all))
assert len(x_all) == 96
assert len(y_all) == 96
if shuffle:
assert not np.allclose(x_all, np.arange(96)[:, None])
assert not np.allclose(y_all, np.arange(1, 97)[:, None])
else:
assert np.allclose(x_all, np.arange(96)[:, None])
assert np.allclose(y_all, np.arange(1, 97)[:, None])
def test_minibatch_generator(shuffle):
data = {"a": np.arange(100)[:, None], "b": np.arange(1, 101)[:, None]}
x_all = []
y_all = []
for _, d in utils.minibatch_generator(data, 10, shuffle=shuffle):
x_all += [d["a"]]
y_all += [d["b"]]
x_all = np.concatenate(x_all)
y_all = np.concatenate(y_all)
if shuffle:
assert not np.allclose(x_all, data["a"])
assert not np.allclose(y_all, data["b"])
x_all = np.sort(x_all, axis=0)
y_all = np.sort(y_all, axis=0)
assert np.allclose(x_all, data["a"])
assert np.allclose(y_all, data["b"])
x_all = []
y_all = []
with pytest.warns(UserWarning):
for _, d in utils.minibatch_generator(data, 12, shuffle=shuffle):
assert d["a"].shape[0] == 12
assert d["b"].shape[0] == 12
x_all += [d["a"]]
y_all += [d["b"]]
x_all = np.sort(np.concatenate(x_all))
y_all = np.sort(np.concatenate(y_all))
assert len(x_all) == 96
assert len(y_all) == 96
if shuffle:
assert not np.allclose(x_all, np.arange(96)[:, None])
assert not np.allclose(y_all, np.arange(1, 97)[:, None])
else:
assert np.allclose(x_all, np.arange(96)[:, None])
assert np.allclose(y_all, np.arange(1, 97)[:, None])
def test_truncation(truncation):
data = {"a": np.random.randn(2, 10), "b": np.random.randn(2, 10)}
duration = 10 if truncation is None else truncation
with pytest.warns(None) as w:
for i, (o, d) in enumerate(utils.minibatch_generator(
data, 2, shuffle=False, truncation=truncation)):
assert np.allclose(d["a"], data["a"][:, o:o + duration])
assert np.allclose(d["b"], data["b"][:, o:o + duration])
# pylint: disable=undefined-loop-variable
assert i == 10 // duration - (10 % duration == 0)
assert len(w) == (1 if truncation == 3 else 0)
# check truncation with n_steps input
for i, (o, d) in enumerate(utils.minibatch_generator(
10, None, truncation=3)):
assert o == i * 3
if i < 3:
assert d == 3
else:
assert d == 1
def test_truncation(truncation):
inputs = {"a": np.random.randn(2, 10)}
targets = {"b": np.random.randn(2, 10)}
duration = 10 if truncation is None else truncation
with pytest.warns(None) as w:
for i, (o, x, y) in enumerate(
utils.minibatch_generator(inputs,
targets,
2,
shuffle=False,
truncation=truncation)):
assert np.allclose(x["a"], inputs["a"][:, o:o + duration])
assert np.allclose(y["b"], targets["b"][:, o:o + duration])
# pylint: disable=undefined-loop-variable
assert i == 10 // duration - (10 % duration == 0)
assert len(w) == (1 if truncation == 3 else 0)