Continue hunting for solution to numpy left-bound.

src/spikeinterface/postprocessing/correlograms.py

@@ -404,8 +404,13 @@ def correlogram_for_one_segment(spike_times, spike_labels, window_size, bin_size
             if sign == -1:
                 mask[:-shift][spike_diff_b < -num_half_bins] = False
             else:
-                spike_diff_b[np.where(spike_diff_b == num_half_bins)] -= 1
+                # spike_diff_b[np.where(spike_diff_b == num_half_bins)] -= 1  adds to the first AND last bin, which we dont want.
                 mask[:-shift][spike_diff_b >= num_half_bins] = False
+                # spike_diff_b[spike_diff_b == num_half_bins] = 0  # fills the central bin
+                # the problem is that we need to mask specific pairs of comparisons
+                # but this is just masking the entire spike time.
+                # I still don't understand why removing the bound is leading to error at all,
+                # it is leading to extra counting.
 
             m = mask[:-shift]
 

src/spikeinterface/postprocessing/tests/test_correlograms.py

@@ -197,6 +197,9 @@ def test_detect_injected_correlation():
 # 3) some different window_ms and bin_ms. Instead, maybe core utils
 # can be factored out and segment stuff is handled in a new test.
 # otherwise, will be too much!
+
+
+# keep the window and bin fixed. test across num filled bins
 def test_correlograms_unit():
     """ """
     sampling_frequency = 30000
@@ -242,9 +245,11 @@ def test_correlograms_unit():
 
     # TODO: tidy up this test, add multi-segment case. Decide which cases to test neatly.
     for auto_idx in [(0, 0), (1, 1)]:
-        breakpoint()
-        assert np.array_equal(expected_results, result_numpy[auto_idx])  # TODO: CHECK!
-        assert np.array_equal(expected_results, result_numba[auto_idx])
+        try:
+            assert np.array_equal(expected_results, result_numpy[auto_idx])  # TODO: CHECK!
+            assert np.array_equal(expected_results, result_numba[auto_idx])
+        except:
+            breakpoint()
 
     expected_results[int(num_bins / 2)] = num_filled_bins
     for auto_idx in [(1, 0), (0, 1)]: