new feature peak-to-peak variation

light-curve/light_curve/light_curve_py/__init__.py

@@ -24,6 +24,7 @@
 from .features.otsusplit import *
 from .features.pdiffmperc import *
 from .features.percampl import *
+from .features.ptp_var import *
 from .features.rainbow import *
 from .features.redchi2 import *
 from .features.roms import *

light-curve/light_curve/light_curve_py/features/ptp_var.py

@@ -0,0 +1,33 @@
+import numpy as np
+
+from ._base import BaseSingleBandFeature
+
+
+class PeakToPeakVar(BaseSingleBandFeature):
+    r"""Peak-to-peak variation
+ 
+    $$
+    \frac{(m_i - \sigma_i)_\text{max} - (m_i + \sigma_i)_\text{min}}{(m_i - \sigma_i)_\text{max} + (m_i + \sigma_i)_\text{min}}
+    $$
+    For non-variable data, it should be close to zero.
+    If data is close to be variable, the index should be more or equal than 0.10-0.15.
+    It is sensitive to magnitude of error values and can be negative in overestimated errors case.
+    
+    - Depends on: **flux density**, **errors**
+    - Minimum number of observations: **2**
+    - Number of features: **1**
+
+    Aller M.F., Aller H.D., Hughes P.A. 1992. [DOI:10.1086/171898](https://www.doi.org/10.1086/171898)
+    """
+
+    def _eval_single_band(self, t, m, sigma=None):
+        a = np.max(m - sigma)
+        b = np.min(m + sigma)
+        return (a - b) / (a + b)
+
+    @property
+    def size_single_band(self):
+        return 1
+
+
+__all__ = ("PeakToPeakVar",)

light-curve/tests/light_curve_py/features/test_ptp_var.py

@@ -0,0 +1,74 @@
+import numpy as np
+from numpy.testing import assert_allclose
+
+from light_curve.light_curve_py import PeakToPeakVar
+
+
+def test_ptpvar_const_data():
+    feature = PeakToPeakVar()
+    n = 100
+    t = np.arange(n)
+    m = np.ones_like(t)
+    sigma = 0.1 * np.ones_like(t)
+    actual = feature(t, m, sigma)
+    desired = -0.1
+    assert_allclose(actual, desired)
+
+
+def test_ptpvar_periodic_data():
+    feature = PeakToPeakVar()
+    n = 100
+    t = np.linspace(0, np.pi, n)
+    m = np.sin(t)
+    sigma = 0.1 * np.ones_like(t)
+    actual = feature(t, m, sigma)
+    desired = 0.8
+    assert_allclose(actual, desired, rtol=3 / np.sqrt(n))
+
+
+def test_ptpvar_norm_data_1():
+    rng = np.random.default_rng(0)
+    n = 100
+    t = np.linspace(0, 1, n)
+    m = np.abs(rng.normal(0, 1, n))
+    sigma = 0.2 * np.ones_like(t)
+    feature = PeakToPeakVar()
+    actual = feature(t, m, sigma)
+    desired = 1
+    assert_allclose(actual, desired, rtol=3 / np.sqrt(n))
+
+
+def test_ptpvar_norm_data_2():
+    rng = np.random.default_rng(0)
+    n = 10000
+    t = np.linspace(0, 1, n)
+    m = np.abs(rng.normal(0, 1, n))
+    sigma = 0.2 * np.ones_like(t)
+    feature = PeakToPeakVar()
+    actual = feature(t, m, sigma)
+    desired = 1
+    assert_allclose(actual, desired, rtol=3 / np.sqrt(n))
+
+
+def test_ptpvar_expon_data_1():
+    rng = np.random.default_rng(0)
+    n = 100
+    t = np.linspace(0, 1, n)
+    m = rng.exponential(2, n)
+    sigma = np.ones_like(t)
+    feature = PeakToPeakVar()
+    actual = feature(t, m, sigma)
+    desired = 1
+    assert_allclose(actual, desired, rtol=3 / np.sqrt(n))
+
+
+def test_ptpvar_expon_data_2():
+    rng = np.random.default_rng(0)
+    n = 10000
+    t = np.linspace(0, 1, n)
+    m = rng.exponential(2, n)
+    sigma = np.ones_like(t)
+    feature = PeakToPeakVar()
+    actual = feature(t, m, sigma)
+    desired = 1
+    assert_allclose(actual, desired, rtol=3 / np.sqrt(n))