Update polyfit to interpolation

ETA.py

@@ -8,6 +8,7 @@ import shelve
 from scipy.signal import convolve as sp_convolve
 from scipy.signal import find_peaks
 from scipy.ndimage import rotate
+import scipy.interpolate as interpolate
 
 cache , filename , output , calibration = None , None , None , None
 verbose , no_cache = False , False
@@ -478,12 +479,16 @@ if calibration != None:
     peaks_calib = np.loadtxt( calibration ) # sorted list
     peaks_calib = np.sort( peaks_calib )
     mean_up   = np.mean( data[
-        calibrations[ 'top' ][ 'y' ][ 'min' ] : calibrations[ 'top' ][ 'y' ][ 'max' ],
+        calibrations[ 'top' ][ 'y' ][ 'min' ]:
-        calibrations[ 'top' ][ 'x' ][ 'min' ] : calibrations[ 'top' ][ 'x' ][ 'max' ]
+        calibrations[ 'top' ][ 'y' ][ 'max' ],
+        calibrations[ 'top' ][ 'x' ][ 'min' ]:
+        calibrations[ 'top' ][ 'x' ][ 'max' ]
     ] , axis = 0 )
     mean_down = np.mean( data[
-        calibrations[ 'down' ][ 'y' ][ 'min' ] : calibrations[ 'down' ][ 'y' ][ 'max' ],
+        calibrations[ 'down' ][ 'y' ][ 'min' ]:
-        calibrations[ 'down' ][ 'x' ][ 'min' ] : calibrations[ 'down' ][ 'x' ][ 'max' ]
+        calibrations[ 'down' ][ 'y' ][ 'max' ],
+        calibrations[ 'down' ][ 'x' ][ 'min' ]:
+        calibrations[ 'down' ][ 'x' ][ 'max' ]
     ] , axis = 0 )
     peaks_up   = np.array(
         utils.retrieve_peaks(
@@ -559,7 +564,9 @@ if calibration != None:
         cons_diff , i = utils.same_value( diff ) , 0
         for list_same in cons_diff:
             new_data[ i : i + len( list_same ) , x ] = data[
-                calibrations[ 'top' ][ 'y' ][ 'max' ] + i : calibrations[ 'top' ][ 'y' ][ 'max' ] + i + len( list_same ),
+                calibrations[ 'top' ][ 'y' ][ 'max' ] + i :
+                calibrations[ 'top' ][ 'y' ][ 'max' ] + i
+                + len( list_same ),
                 calibrations[ 'top' ][ 'x' ][ 'min' ] + list_same[0]
             ]
             i += len( list_same )
@@ -600,11 +607,13 @@ if verbose:
 """
 remove artefact
 """
-list_ = np.convolve(
+first_stripe = np.argmax(
+    np.convolve(
         np.gradient(
             np.mean(
                 data[
-                calibrations[ 'top' ][ 'y' ][ 'max' ] : calibrations[ 'down' ][ 'y' ][ 'min' ],
+                    calibrations[ 'top' ][ 'y' ][ 'max' ] :
+                    calibrations[ 'down' ][ 'y' ][ 'min' ],
                     border[ 'x' ][ 'min' ] : border[ 'x' ][ 'max' ]
                 ]       ,
                 axis = 1,
@@ -612,26 +621,74 @@ list_ = np.convolve(
         )            ,
         np.ones( 50 ),
         'same'       ,
+    ),
 )
-first_stripe = np.argmax( list_ )
+range_       = np.arange(
-range_       = range(
     calibrations[ 'top'  ][ 'y' ][ 'max' ] + first_stripe,
-    calibrations[ 'down' ][ 'y' ][ 'max' ]
+    calibrations[ 'down' ][ 'y' ][ 'max' ]               ,
+    100                                                  ,
 )
 
 for line in range_:
-    abciss = range(
+    convolve_point = 500
+    abciss = np.arange(
         border[ 'x' ][ 'min' ],
         border[ 'x' ][ 'max' ],
     )
-    list_  = data[
+    list_  = np.convolve(
+        data[
             line                                           ,
             border[ 'x' ][ 'min' ] : border[ 'x' ][ 'max' ]
-    ]
+        ]      ,
-    polyfit_ = np.polyfit(
+        np.ones(
-        abciss,
+            convolve_point,
+        )      ,
+        'valid', # Need to append value at extremities
+    ) / convolve_point # keep correct value
+    list_ = np.concatenate(
+        (
+            data[
+                line                                        ,
+                border[ 'x' ][ 'min' ]                      :
+                border[ 'x' ][ 'min' ] + convolve_point //
+                2 - 1
+            ]    ,
             list_,
-        8     ,
+            data[
+                line                                                ,
+                border[ 'x' ][ 'max' ]
+                - convolve_point // 2 + convolve_point % 2          :
+                border[ 'x' ][ 'max' ]
+            ]    ,
+        ),
+    )
+    number_point  = 10
+    points_taken  = abciss[
+        convolve_point // 2 + convolve_point % 2  :
+        - convolve_point // 2 + convolve_point % 2:
+        ( abciss.shape[0] - convolve_point + 1 ) // ( number_point - 1 )
+    ]
+    values_taken  = list_[
+        convolve_point // 2 + convolve_point % 2  :
+        - convolve_point // 2 + convolve_point % 2:
+        ( abciss.shape[0] - convolve_point + 1 ) // ( number_point - 1 )
+    ]
+    sigma         = np.std( list_ )
+    interpolation = interpolate.PchipInterpolator( # Seems the best
+                    # (Have to check to be sure) -> TODO find reference
+                    # value to determine which interpolation is better
+        points_taken,
+        values_taken,
+    )
+    wrong_points = np.where(
+        np.abs(
+            interpolation(
+                abciss,
+            ) - list_ ,
+        ) > sigma
+    )[0]
+    list_[ wrong_points ] = interpolation( # no border point
+        abciss[ wrong_points ],
     )
 
     import matplotlib.pyplot as plt
@@ -641,8 +698,7 @@ for line in range_:
     )
     plt.plot(
         abciss,
-        np.polyval(
+        interpolation(
-            polyfit_,
             abciss,
         )     ,
     )