weixin_39843151
weixin_39843151
2020-12-08 18:55

Cov channels

the changes made recently in terms of excluding bad channels have caused a regression in covariance calculation.

Due to how inverse calculations are coded (and for backward compatibility reasons), Covariance matrices must include data from all channels, including bad ones. The current version of master doesn't use all channels in cov calculation, though, since including the bad channels in epoching will cause epochs to be dropped because of the bad channels. In other words, compute_covariance requires passing an epochs object with all channels present, including the channels marked bad. However, generating an epochs instance with these bad channels included will unnecessarily drop epochs.

One possible solution is to never drop epochs on the basis of bad channels, even if those bad channels are requested in the epochs construction. To avoid user confusion, I think we should add this functionality (remember it was one of the proposed solutions anyway), and then default to keeping all channels (including bad ones) in Epoch generation. That way, a user can process raw, generate epochs with no picks argument, and have this generate a correct, useable Covariance. Make sense?

该提问来源于开源项目:mne-tools/mne-python

  • 点赞
  • 写回答
  • 关注问题
  • 收藏
  • 复制链接分享
  • 邀请回答

4条回答

  • weixin_39885412 weixin_39885412 4月前

    +1 for default to keep all channels in Epochs if no picks are provided.

    点赞 评论 复制链接分享
  • weixin_39843151 weixin_39843151 4月前

    ...and don't DQ trials based on bad channels, I presume you're also okay with. It'll require modifying _is_good(), but that's not so terrible.

    点赞 评论 复制链接分享
  • weixin_39885412 weixin_39885412 4月前

    DQ?

    we should drop epochs excluding bad channels yes.

    点赞 评论 复制链接分享
  • weixin_39843151 weixin_39843151 4月前

    sorry, disqualify :)

    点赞 评论 复制链接分享

相关推荐