Eric Larson comments

Results 1498 comments of


                                            Eric Larson

Use np.linalg.multi_dot instead of multiple np.dot routines

AFAIK there isn't a mechanism for the `__matmul__` / `@` operator to modify the order of operations if it's chained, so it's very likely multiple `@` will be slower than...

Use np.linalg.multi_dot instead of multiple np.dot routines

Makes sense. Looks like there is ~30 uS overhead for deciding which order to use, so we'd only want to think about this if the sizes are big enough. More...

Use np.linalg.multi_dot instead of multiple np.dot routines

Do you need to update `sample` or `testing`? Something like this should work ``` mne.datasets.sample.data_path(force_update=True, download=True, verbose=True) ```

Use np.linalg.multi_dot instead of multiple np.dot routines

You need to `pip install pytest-cov` or `conda install pytest-cov`

Use np.linalg.multi_dot instead of multiple np.dot routines

> As another point, using the np.dot is preferable if you are certain of the matrix forms and the best ordering. If you are not, it is recommended that you...

Use np.linalg.multi_dot instead of multiple np.dot routines

> we don't always know the dimensions in advance, but in an N×M @ M×M @ M×P operation we usually know that, for example, N

Use np.linalg.multi_dot instead of multiple np.dot routines

It's not really large vs small, it's tall vs short. If everything is square it doesn't matter. Imagine an operation of the following, using shapes in place of var names:...

ENH: Improve representation of epochs time bar

BTW my vote would be to display anything < 1 s using `x ms` and >= 1 s using `x s`, I think it'll be easier for people to read...

ENH: Split whats_new

That would be fine, too!

[DOC] Update description of 'epoch_threshold' in compute_bridged_electrodes

Thanks @mscheltienne