Describe the new feature or enhancement

Discussed partly with @larsoner.

MNE-Connectivity is being updated to allow connectivity computations using complex coefficients stored in EpochsTFR objects (mne-tools/mne-connectivity#232). For the multitaper method, TFR coefficients have a tapers dimension that need to be aggregated over when computing the CSD. These weights are not stored in the *TFR objects, and computing these weights requires knowledge of the time_bandwidth and n_cycles used to estimate the TFR. However, currently these parameters are not stored in *TFR objects.

If we want to accurately compute these weights in MNE-Connectivity, we would currently require the user to pass these parameters in addition to the EpochsTFR object. It would be more convenient if this information could also be read from the EpochsTFR object.

Describe your proposed implementation

There were two ideas:

• Either store this information directly as attributes, e.g., tfr.n_cycles and tfr.time_bandwidth
• Or store these params as a dict to minimise clutter in the *TFR namespace, e.g., tfr.estimation_parameters = {'mt_bandwidth': 4, 'n_cycles': 5}

Describe possible alternatives

An alternative would be to just store the taper weights in the *TFR objects, which is similar to how this is now handled for *Spectrum objects. However, the weights are not returned by tfr_array_multitaper, so they would still need to be recomputed.

It's perhaps simpler if we just handle this in MNE-Connectivity using the stored estimation params.

What do people think is the best choice?