Base knife-edge scan analysis implementation

Hi @lleguy @mercadil,

This is an example of a base implementation, specifically the knife edge scan analysis from Laurent's tb.knife_edge. This is agnostic from xarray data format, thus this can be easily used on other means, such as on a Karabo device.

I also have found out that there could be problems with using bisect especially with repetitive elements. I use slicing instead to find the valid range.

src/toolbox_scs/base/tests/test_knife_edge.py

 import numpy as np
 import pytest
 
-from ..knife_edge import prepare_arrays, range_mask
+from ..knife_edge import erfc, knife_edge_base, prepare_arrays, range_mask
 
 
 def test_range_mask():
@@ -58,7 +58,7 @@ def test_prepare_arrays_nans():
     assert intensities.shape == (size,)
 
     # Test finite motors and signals with NaNs
-    signal_nan = with_values(signal, value=np.nan, num=20)
+    signal_nan = _with_values(signal, value=np.nan, num=20)
     positions, intensities = prepare_arrays(motor, signal_nan)
     assert positions.shape == (size-20,)
     assert np.isfinite(positions).all()
@@ -66,7 +66,7 @@ def test_prepare_arrays_nans():
     assert np.isfinite(intensities).all()
 
     # Test finite signals and motors with NaNs
-    motor_nan = with_values(motor, value=np.nan, num=3)
+    motor_nan = _with_values(motor, value=np.nan, num=3)
     positions, intensities = prepare_arrays(motor_nan, signal)
     assert positions.shape == ((trains-3) * pulses,)
     assert np.isfinite(positions).all()
@@ -91,7 +91,23 @@ def test_prepare_arrays_size():
     assert intensities.shape == (size,)
 
 
-def with_values(array, value, num=5):
+def test_knife_edge_base():
+    p0 = [0, -1.5, 1, 0]
+    x = np.linspace(-3, 3)
+    y = erfc(x, *p0)
+    noise = y * np.random.normal(0, .02, y.shape)  # 2% error
+    eff_y = (y + noise).reshape(1, -1)
+
+    # Test noisy data
+    popt, _ = knife_edge_base(x, eff_y)
+    np.testing.assert_allclose(p0, popt, atol=1e-1)
+
+    # Test flipped data
+    popt, _ = knife_edge_base(x, eff_y[::-1])
+    np.testing.assert_allclose(p0, popt, atol=1e-1)
+
+
+def _with_values(array, value, num=5):
     copy = array.astype(np.float)
     copy.ravel()[np.random.choice(copy.size, num, replace=False)] = value
     return copy