Fix RANSAC-breaking bug in RANSAC correlation window chunking code (#67)

* Fixed reshaping bug in RANSAC

* Updated whats_new.rst

* Improved whats_new.rst

Author: a-hurst

docs/whats_new.rst

@@ -45,6 +45,7 @@ Bug
 - Fixed RANSAC to give consistent results with a fixed seed across different chunk sizes, by `Austin Hurst`_ and `Yorguin Mantilla`_ (:gh:`43`)
 - Fixed "bad channel by flat" threshold in :meth:`NoisyChannels.find_bad_by_nan_flat` to be consistent with MATLAB PREP, by `Austin Hurst`_ (:gh:`60`)
 - Changed "bad channel by deviation" and "bad channel by correlation" detection code in :class:`NoisyChannels` to compute IQR and quantiles in the same manner as MATLAB, thus producing identical results to MATLAB PREP, by `Austin Hurst`_ (:gh:`57`)
+- Fixed a bug where EEG data was getting reshaped into RANSAC windows incorrectly (channel samples were not sequential), which was causing considerable variability and noise in RANSAC results, by `Austin Hurst`_ (:gh:`67`)
 
 API
 ~~~

pyprep/ransac.py

@@ -273,16 +273,18 @@ def _ransac_correlations(
 
     # For the actual data
     data_window = data[chans_to_predict, : n * w_correlation]
-    data_window = data_window.reshape(len(chans_to_predict), n, w_correlation)
+    data_window = data_window.reshape(len(chans_to_predict), w_correlation, n)
+    data_window = data_window.swapaxes(1, 0)
 
     # For the ransac predicted eeg
     pred_window = ransac_eeg[: len(chans_to_predict), : n * w_correlation]
-    pred_window = pred_window.reshape(len(chans_to_predict), n, w_correlation)
+    pred_window = pred_window.reshape(len(chans_to_predict), w_correlation, n)
+    pred_window = pred_window.swapaxes(1, 0)
 
     # Perform correlations
     for k in range(w_correlation):
-        data_portion = data_window[:, :, k]
-        pred_portion = pred_window[:, :, k]
+        data_portion = data_window[k, :, :]
+        pred_portion = pred_window[k, :, :]
 
         R = np.corrcoef(data_portion, pred_portion)