MAINT: fix KF prediction

SiLab-Bonn · Jul 25, 2017 · 545147e · 545147e
1 parent 4c99c85
commit 545147e
Show file tree

Hide file tree

Showing 3 changed files with 16 additions and 16 deletions.
diff --git a/testbeam_analysis/testing/test_kalman.py b/testbeam_analysis/testing/test_kalman.py
@@ -54,16 +54,16 @@ def test_kalman(self):
                   'add_scattering_plane': False}
 
         # expected result array: (state estimates, chi, x error, y errors)
-        result = np.array([[[-1.23045812e+03, 2.82684464e+03, 9.54189393e-04, 5.78723069e-05],
-                            [-1.25900270e+03, 2.82511339e+03, 9.54667995e-04, 5.79013344e-05],
-                            [-1.28705254e+03, 2.82443254e+03, 9.22692240e-04, 2.23966275e-05],
-                            [-1.30575083e+03, 2.82550588e+03, 8.57719412e-04, -4.92360235e-05],
-                            [-1.33339390e+03, 2.83014572e+03, 7.55275169e-04, -1.26771525e-04],
-                            [-1.36192826e+03, 2.83782855e+03, 6.79389545e-04, -1.82924543e-04],
-                            [-1.38713361e+03, 2.84461505e+03, 6.79389545e-04, -1.82924543e-04]],
-                           [74],
-                           [3.62429044, 3.2884327, 3.21655702, 3.1539946, 3.23671172, 4.66501707, 8.62909928],
-                           [3.62429044, 3.2884327, 3.21655702, 3.1539946, 3.23671172, 4.66501707, 8.62909928]])
+        result = np.array([[[-1.23001328e+03, 2.82727830e+03, -9.15880110e-04, -2.09034100e-05],
+                            [-1.25862573e+03, 2.82521051e+03, -8.74822163e-04, 2.73502359e-05],
+                            [-1.28712065e+03, 2.82381689e+03, -8.12311281e-04, 1.00543040e-04],
+                            [-1.30702210e+03, 2.82479582e+03, -7.11712303e-04, 1.56181457e-04],
+                            [-1.33425505e+03, 2.82962871e+03, -6.79354825e-04, 1.80316681e-04],
+                            [-1.36061740e+03, 2.83880395e+03, -7.31034579e-04, 1.42175167e-04],
+                            [-1.38773879e+03, 2.84407865e+03, -7.31034579e-04, 1.42175167e-04]],
+                           [37],
+                           [3.67202783, 4.76060863, 3.73543952, 3.67302492, 3.99327848, 4.88787552, 10.82919706],
+                           [3.67202783, 4.76060863, 3.73543952, 3.67302492, 3.99327848, 4.88787552, 10.82919706]])
 
         for i in range(4):  # test each return (state estimates, chi, x error, y errors) seperatly
             test = test_tools._call_function_with_args(function=track_analysis._fit_tracks_kalman_loop,

diff --git a/testbeam_analysis/tools/kalman.py b/testbeam_analysis/tools/kalman.py
@@ -211,8 +211,8 @@ def _filter(alignment, transition_matrices, observation_matrices, transition_cov
                 z_diff = offsets_rotated[1][:, 2] - offsets_rotated[0][:, 2]
 
                 # update transition matrix, only need to change these value in case for rotated planes
-                transition_matrices[:, t - 1, 0, 2] = - z_diff
-                transition_matrices[:, t - 1, 1, 3] = - z_diff
+                transition_matrices[:, t - 1, 0, 2] = z_diff
+                transition_matrices[:, t - 1, 1, 3] = z_diff
 
             # store updated transition matrix for smoothing function
             transition_matrices_update[:, t - 1] = transition_matrices[:, t - 1]

diff --git a/testbeam_analysis/track_analysis.py b/testbeam_analysis/track_analysis.py
@@ -1243,8 +1243,8 @@ def _fit_tracks_kalman_loop(track_hits, dut_fit_selection, pixel_size, n_pixels,
             # If first dut is used in track building, take first dut hit as initial value and
             # its corresponding cluster position error as the error on the measurement.
             initial_state_mean[index] = np.array([actual_hits[0, 0], actual_hits[0, 1], 0., 0.])
-            initial_state_covariance[index, 0, 0] = np.square(x_err[0])  # np.square(pixel_resolution[0, 0])  # np.square(x_err[0])
-            initial_state_covariance[index, 1, 1] = np.square(y_err[0])  # np.square(pixel_resolution[0, 1])  # np.square(y_err[0])
+            initial_state_covariance[index, 0, 0] = np.square(x_err[0])
+            initial_state_covariance[index, 1, 1] = np.square(y_err[0])
         else:  # first dut is not in fit selction
             # Take hit from first dut which is in fit selection. Cannot take hit from first dut,
             # since do not want to pass measurement to kalman filter (unbiased).
@@ -1260,9 +1260,9 @@ def _fit_tracks_kalman_loop(track_hits, dut_fit_selection, pixel_size, n_pixels,
         # rotations of planes into account.
         transition_matrix[index, sel, :, 0] = np.array([1., 0., 0., 0.])
         transition_matrix[index, sel, :, 1] = np.array([0., 1., 0., 0.])
-        transition_matrix[index, sel, :, 2] = np.array([-(z_diff), np.zeros((len(sel),)),
+        transition_matrix[index, sel, :, 2] = np.array([(z_diff), np.zeros((len(sel),)),
                                                         np.ones((len(sel),)), np.zeros((len(sel),))]).T
-        transition_matrix[index, sel, :, 3] = np.array([np.zeros((len(sel),)), -(z_diff),
+        transition_matrix[index, sel, :, 3] = np.array([np.zeros((len(sel),)), (z_diff),
                                                         np.zeros((len(sel),)), np.ones((len(sel),))]).T
 
         # express transition covariance matrices, according to http://web-docs.gsi.de/~ikisel/reco/Methods/CovarianceMatrices-NIMA329-1993.pdf