def test_linearity_mul(self, n, window, mode, const):
np.random.seed(42)
a0 = np.random.randn(10, 20)
a1 = const * a0
diff0 = fdiff(a0, n, window=window, mode=mode)
diff1 = fdiff(a1, n, window=window, mode=mode)
assert_allclose(const * diff0, diff1)
def test_linearity_add(self, n, window, mode):
np.random.seed(42)
a0 = np.random.randn(10, 20)
a1 = np.random.randn(10, 20)
a2 = a0 + a1
diff0 = fdiff(a0, n, window=window, mode=mode)
diff1 = fdiff(a1, n, window=window, mode=mode)
diff2 = fdiff(a2, n, window=window, mode=mode)
assert_allclose(diff0 + diff1, diff2)
def test_mode(self, n, window):
np.random.seed(42)
a = np.random.randn(10, 20)
out_f = fdiff(a, n, axis=0, window=window, mode="full")
out_v = fdiff(a, n, axis=0, window=window, mode="valid")
assert_array_equal(out_f[window - 1:, :], out_v)
out_f = fdiff(a, n, axis=1, window=window, mode="full")
out_v = fdiff(a, n, axis=1, window=window, mode="valid")
assert_array_equal(out_f[:, window - 1:], out_v)
def test_integer_n(self, n, ndim):
"""
fdiff with integer n is equal to numpy.diff
"""
np.random.seed(42)
a = np.random.randn(*([100] * ndim))
for axis in range(ndim):
out = np.diff(a, n=int(n), axis=axis)
assert_array_equal(fdiff(a, n=n, axis=axis), out)
def test_one(self, n, window, n_blanks_1, n_blanks_2, n_features):
a = np.concatenate((
np.zeros((window + n_blanks_1, n_features)),
np.ones((1, n_features)),
np.zeros((window + n_blanks_2, n_features)),
))
coef = fdiff_coef(n, window)
diff = fdiff(a, n=n, window=window, axis=0)
for i in range(n_features):
assert_array_equal(diff[window + n_blanks_1:, i][:window], coef)
assert_equal(diff[:window + n_blanks_1], 0)
assert_equal(diff[window + n_blanks_1 + window:], 0)
def test_append(self, n, window, mode):
a = np.random.randn(10, 20)
append = np.random.randn(5, 20)
cat = np.concatenate((a, append), axis=0)
out = fdiff(cat, n, axis=0, window=window, mode=mode)
result = fdiff(a, n, axis=0, window=window, append=append, mode=mode)
assert_array_equal(result, out)
append = np.random.randn(10, 5)
cat = np.concatenate((a, append), axis=1)
out = fdiff(cat, n, axis=1, window=window, mode=mode)
result = fdiff(a, n, axis=1, window=window, append=append, mode=mode)
assert_array_equal(result, out)
cat = np.concatenate((a, np.zeros((1, 20))), axis=0)
out = fdiff(cat, n, axis=0, window=window, mode=mode)
result = fdiff(a, n, axis=0, window=window, append=0, mode=mode)
assert_array_equal(result, out)
cat = np.concatenate((a, np.zeros((10, 1))), axis=1)
out = fdiff(cat, n, axis=1, window=window, mode=mode)
result = fdiff(a, n, axis=1, window=window, append=0, mode=mode)
assert_array_equal(result, out)
def _add_diff(self, _df, _diff_d, _name, _add=True):
""" 为 数据列 添加 差分数据列
:param _add:
:param _df: 原数据 DataFrame
:param _d_list: 差分阶数 [0.3, 0.5, 0.7]
:param _name: 需要添加 差分值 的数据列 名称
:param _agg_dict:
:param _agg_type:
:param _add:
:return: """
if _add:
if len(_df) >= 12: # 数据行数大于等于12才进行差分操作
_diff_ar = fdiff(_df[_name],
n=_diff_d,
window=10,
mode="valid") # 列差分，不使用未来数据
_paddings = len(_df) - len(_diff_ar) # 差分后数据长度变短，需要在前面填充多少数据
_diff = np.nan_to_num(np.concatenate((np.full(_paddings,
0), _diff_ar)),
nan=0) # 将所有nan替换为0
_df[_name + f'_diff_{_diff_d}'] = _diff # 将差分数据记录到 DataFrame
else:
_df[_name + f'_diff_{_diff_d}'] = np.nan # 数据行数不足12的填充为空数据
def test_transform(self, d, window, mode):
np.random.seed(42)
X = np.random.randn(100, 200)
fracdiff = Fracdiff(d=d, window=window, mode=mode)
out = fdiff(X, n=d, axis=0, window=window, mode=mode)
assert_array_equal(fracdiff.fit_transform(X), out)
def test_axis(self, n):
np.random.seed(42)
a = np.random.randn(10, 20, 30, 40)
out1 = np.swapaxes(fdiff(np.swapaxes(a, 0, 1), n, axis=0), 0, 1)
out2 = np.swapaxes(fdiff(np.swapaxes(a, 0, 2), n, axis=0), 0, 2)
out3 = np.swapaxes(fdiff(np.swapaxes(a, 0, 3), n, axis=0), 0, 3)
assert_array_equal(fdiff(a, n, axis=1), out1)
assert_array_equal(fdiff(a, n, axis=2), out2)
assert_array_equal(fdiff(a, n, axis=3), out3)
assert_array_equal(fdiff(a, n, axis=0), fdiff(a, n, axis=-4))
assert_array_equal(fdiff(a, n, axis=1), fdiff(a, n, axis=-3))
assert_array_equal(fdiff(a, n, axis=2), fdiff(a, n, axis=-2))
assert_array_equal(fdiff(a, n, axis=3), fdiff(a, n, axis=-1))