SpikeInterface
The waveforms are upside down.
I can't find the reason of this problem. The waveforms in the Phy are normal, but the waveforms are upside down when the results are imported into the sorting analyzer.
Could you let us know which recording you are using? Maybe a little bit of the script you ran to make your sorting analyzer. If the gain has the wrong sign this could happen for sure. So it would help us to see what you did to get to this point :)
Yes I think you might not be looking at the same channel. How are you generating the spikeinterface plot?
Sorry. I will add my code here.
First, I export the sorting analyzer.
export_to_phy(sa, phy_res_path, compute_amplitudes=True, compute_pc_features=True, remove_if_exists=True)
Then, I read the data from the result.
info_path = os.path.join(phy_folder, 'reportinfo.json')
file = open(info_path, 'r')
params = json.load(file)
peak_sign = params['peak_sign']
radius_um = params['radius_um']
file_names = params['file_names']
frame_list = params['frame_list']
channel_ids = params['channel_ids']
gain_to_uV = params['gain_to_uV']
file.close()
params = si.read_python(str(phy_folder / "params.py"))
sample_rate = params["sample_rate"]
rec_data_path = params['dat_path']
data_type = params['dtype']
headeroffset = params['offset']
rec_all = si.read_binary(rec_data_path, sampling_frequency=sample_rate, dtype=data_type, file_offset=headeroffset, \
num_channels=len(channel_ids), channel_ids=channel_ids, offset_to_uV=0, gain_to_uV=gain_to_uV, is_filtered=True)
The 'gain_to_uV' in reportinfo.json is 1. And the plotting code is following.
for unit_id in sorted(sorting.unit_ids, reverse=True):
unit_sa = sorting_analyzer.select_units(unit_ids=[unit_id])
wv = unit_sa.get_extension(extension_name="waveforms")
waveforms = wv.get_data()
mean_waveform = np.mean(waveforms[:, :, ch], axis=0)
fig, axes = plt.subplots(nrows=1, ncols=2, width_ratios=[0.7, 0.3])
max_plot_wf = 100 if waveforms.shape[0] > 100 else waveforms.shape[0]
for i in range(max_plot_wf):
axes[0].plot(waveforms[i, :, ch], alpha=0.3, color='grey',linewidth=0.5)
axes[0].plot(mean_waveform, color='black',linewidth=1)
...
In fact, in the units of the sorting result, some are same as the picture in Phy, but some are upside down.
sorting_curated_phy = se.read_phy(phy_folder, exclude_cluster_groups=["noise"])
rec_all = si.read_binary(rec_data_path, sampling_frequency=sample_rate, dtype=data_type, file_offset=headeroffset, \
num_channels=len(channel_ids), channel_ids=channel_ids, offset_to_uV=0, gain_to_uV=gain_to_uV, is_filtered=True)
sorting_analyzer = si.create_sorting_analyzer(sorting=sorting_curated_phy, recording=rec_all, sparse=False)
Sorry is your "binary_file" saved post-filtering + referencing/preprocessing? The SortingAnalyzer performs best with filtered and referenced data. So I want to make sure we are seeing this.
Thank you for your reply. I know it's difficult to find the solution without the data. The binary_file is saved with Phy, so I think the waveforms reconstructed by Phy are same as that I read for plotting. And the data saving process is like this:
sa = si.create_sorting_analyzer(sorting=sorting, recording=rec_correct)
export_to_phy(sa, phy_res_path, compute_amplitudes=True, compute_pc_features=True, remove_if_exists=True)
The "rec_correct" is preprocessed using the pipeline "common_reference" -> "bandpass_filter" -> "correct_motion".
What if you do sparse=True? I'm pretty sure phy still only picks a subset of channels. So if we use our sparsity (which won't be exactly the same set of channels, but will be close) could you try that?
Thank you for your advice. But it seems same as before. I will try to load the recording from sorting analyzer's result, and find the difference between the two methods of saving recording.
I found the reason! My fault. I used the wrong channel id. Thank you for your reply.