SpikeInterface
get_chunk_with_margin() should accept return_scaled
As mentionned in #2346, there is slight problem with the sortingcomponents. Currently, because of the internal function get_chunk_with_margin that do not accept return_scaled, all internal operations are performed under the hood on raw data (no scaling). This is quite often not a major issue, but if some fileformat have a negative gain, then the peaks are all upside down internally for the components. We could simply extend this function (and therefore all the node_pipelines) to accept a return_scaled argument, and leave it as default to True, as much as possible?
@yger this is very tricky. The reason why return_scaled is only handled by the Recording is that if you have a processing chain you only want to apply scaling once!
In my opinion, for the Neuralynx issue, the user should set the sign to positive, or sorters could do that automatically if they detect a negative gain. What do you think?
Hum... I did not consider such a problem. I'll keep thinking about it, but if no better solution, then what you are suggesting is indeed the way to proceed. If a negative gain is detected, then we should flip peak signs internally.
Actually, new proposal before discussion. Since the problem is only in the run_node_pipeline, mostly used in components, then maybe in the node pipeline, when we want to access the (final, i.e. the end of preprocessing chain) traces for the nodes, we could use a different function (get_traces_with_margin) that will make an internal call to get_traces (and thus get the scaling right). So we keep get_chunk_with_margin as a low-level function, not making any rescaling, and we have a new one (only used in the node pipeline) get scaled data. To be discussed